KR20060088134A - Image processing device - Google Patents

Abstract

An image processing method which digitizes similarity between a first pixel and a second pixel constituting an image by means of a statistic test, and averages a first pixel value and a second pixel value when the digitized similarity is high or does not average them when the determined similarity is low, whereby reducing noise without compromise in spatial resolution and in temporal resolution.

Description

Image Processing Equipment {IMAGE PROCESSING DEVICE}

1 is a table showing the relationship between risk (p (x, y)), weight (w (p (x, y))) and null hypothesis,

Fig. 2 is a table showing various application examples of the present embodiment of the coherent filter according to the present invention classified according to the difference in the configuration method of the pixel value v (x);

3 is a diagram showing an example of the schematic configuration of an X-ray CT device incorporating a coherent filter;

4 is an explanatory diagram conceptually illustrating image processing by a coherent filter according to the present invention;

5 to 7 show an example in which noise suppression processing by a coherent filter according to the present invention is performed on a plurality of images constituting a dynamic CT image, and FIG. 5 shows an unprocessed original image, FIGS. 6 and 7. Are diagrams each showing an image of processing completion,

8 is a flowchart showing a flow of noise suppression processing by the coherent filter of the present embodiment;

9 to 11 show an example in which noise suppression processing by a coherent filter according to the present invention is performed on one CT image, FIG. 9 shows an unprocessed original image, and FIG. 10 shows an image of processing completed. 11 represents the image which averaged the image shown in FIG. 9, and the image shown in FIG. 10, respectively.

12 to 15 are contrast images showing 3D extraction examples of blood vessels, and FIG. 16 is a view showing images of brain proteins extracted as another extraction example.

17 is an explanatory diagram showing an example of a method of constructing a pixel value v (x) with one image;

18 is an explanatory diagram showing another example of a method of constructing a pixel value v (x) with one image;

19 is a view showing a time concentration curve when the coherent regression method and the AUC method are combined, and the same curve when the conventional method and the AUC method are combined;

FIG. 20 relates to another estimation method of variance σ ² , such as Equation 4 in the specification text, and is an explanatory diagram illustrating a method of calculating variance σ ^{2 ++} by the other estimation method.

It is explanatory drawing explaining the "smoothing process" which is a conventional noise suppression method.

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus, and more particularly, to an image processing apparatus which reduces noise present on an image.

At present, image processing technology is intended to be used in various fields.

Image processing is carried out to cope with deterioration or modification of an image obtained from a videotape recorder or digital camera, or to clearly understand the pattern or structure of the structure to check whether the structure is manufactured as designed. It is carried out for the purpose.

Various image processing is also performed in various medical image diagnosis apparatuses such as an X-ray CT apparatus, a SPECT apparatus, and an MRI apparatus. The utility of the extraction of blood streams and contrast agents, extraction of lesions, contour extraction of organs, etc. is widely recognized.

The image processing technique consists of various element techniques such as noise suppression technique, feature extraction technique, pattern recognition technique, and is used alone or in combination as appropriate. In particular, among such element technologies, a technique for reducing random noise included in an image is indispensable in order to more clearly reproduce an object that has been picked up or reconstructed.

However, further improvement is required in conventional image processing technology, that is, noise reduction technology. For example, so-called "smoothing" is widely known as a noise reduction technique. This smoothing means that when there is an input value (f (i, j)) for an arbitrary pixel (i, j), the average value near the pixel (i, j) is output value for the pixel (i, j). It is set to (g (i, j)). Specifically, assuming that n × n pixels near the pixels i and j are used, the output value g (i, j) is

[Equation 1]

Obtained as

However, a, b, c, and d in Equation 1 are integers. In addition, 1 / (b-a + 1) (d-c + 1) in the said Formula (1) is called a weight. In addition, FIG. 21 has shown the case where a, b, c, d = -1, 1, -1, 1. FIG.

By the way, in general, it is known that if the mean value of n samples taken independently from the distribution of the population whose variance is sigma ² is calculated, the variance of the mean value is sigma ² / n. And "the variance (σ ² )" correspond to the probability distribution using the component attributable to noise included in the value of each pixel (i, j) as the random variable and the variance, so that the value of each pixel (f ( The contribution of noise of i, j)) can be reduced.

However, only by applying this alone, so-called "edge blurriness" occurs, the spatial resolution of the image is impaired, and the whole becomes dim. Taking the above medical image as an example, even when it is desired to describe the fine blood vessel structure with as little noise as possible, according to the noise suppression process according to Equation 1 above, averaging (smoothing) is performed including pixels that do not originally describe the blood vessel structure. Therefore, even when noise is suppressed, the contrast representing the vascular structure is also reduced by smoothing, which makes it difficult to describe the detailed vascular structure.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and an object of the present invention is to effectively suppress noise without causing blurring of an image, and to effectively perform other image processing techniques, for example, a pattern recognition technique. It is to provide an image processing apparatus that can contribute.

(Basic configuration of the present invention)

The present invention first constructs a pixel value that is a vector value or a scalar value for a pixel constituting an image, quantifies a goodness of fit between the pixel value and another pixel value configured separately from each pixel value, When the pixel value is constructed using the other pixel value, if the goodness of fit is large, the contribution of the other pixel is increased, and if the goodness of fit is small, the contribution of the other pixel is made small to constitute the new pixel value. An image processing apparatus characterized by the above-mentioned.

Moreover, this invention may be comprised based on the other pixel which the said "other pixel value comprised separately" comprises the said image. That is, in this case, the pixel value is configured as "one pixel value" and "other pixel value" which are vector values or scalar values for one pixel and another pixel, respectively.

In addition, the present invention particularly applies a weighting function that is a function of the goodness of fit to the goodness of fit obtained for each of the other pixel values to determine the respective weights of the other pixel values, and subsequently the other pixel values using this weight. Computing the new pixel value by calculating the weighted average of increases the weighted average of the other pixel values by increasing the weight when the goodness of fit is large, and when the goodness of fit is small, By reducing the weight, the contribution of the weighted average of the other pixel values is reduced to form the new pixel value.

According to this process, one pixel and another pixel determined to be "similar" are regarded as important, and a new pixel value is constructed, thereby impairing spatial resolution as in the prior art, or impairing temporal resolution when the image is a moving picture. There is nothing to let you do.

(Weighted functions and weights)

In the present invention, it is preferable that the weight function is a nonnegative monotonic increasing function related to the goodness of fit.

And x and y for the one pixel and the other pixel, respectively, and v (x) = (v ₁ (x), v ₂ (x), ... v _K ( x)) and v (y) = (v ₁ (y), v ₂ (y), ..., v _K (y)), and the new pixel value to be configured is v '(x) = (v ′ ₁ (x), v ′ ₂ (x),…, v ′ _K (x)), the fitness is ρ (x, y), the weighting function is w, and the weighting function w is applied to the goodness of fit. When the weight obtained by making the weight w (ρ (x, y)) is obtained,

[Equation 2]

(N (x) represents a range in which the other pixel is included.)

(Hereinafter, a form such as the right side of Equation 2 for obtaining a new pixel value v '(x) is referred to as "coherent filter" according to the present invention.)

(Example of Applicable Image Processing Technology of the Present Invention)

The present invention provides various applications such as reducing noise of the image or performing pattern recognition of the image by configuring the above-described process, that is, the new pixel value on the entire surface of the image, for example. This is possible, and as will be described later, the present invention exhibits excellent effects in each element of the image processing technique.

(Fit)

In the present invention, the fitness may be quantified based on a risk obtained by applying a statistical test method to scalar values constituting the one pixel value and the other pixel value, respectively. In addition, the "statistical test method" said here can be considered, for example, the "x-square test" etc. as mentioned later. In general, the relationship between the risk and the fitness introduced here includes, in particular, the case in which one side increases and the other side decreases. That is, when constructing the new pixel value, the greater the risk factor (when the suitability is smaller), the smaller the weight for the other pixel (the smaller the contribution to the new pixel value to be configured), and the lower the risk factor (the It is a case where the weight for the other pixel becomes larger (contribution to the new pixel value to be configured) becomes larger.

(Selection range of `` other pixels '')

In addition, the present invention may be such that the other pixel is selected in a predetermined area around the one pixel, and in particular, the predetermined area is near the center of the one pixel. With such a configuration, other pixels which are determined to be similar to one pixel are generally more likely to exist around the one pixel, so that when the above-mentioned area is selected, a new pixel value can be used effectively. It is possible to omit a missing operation.

(Composition method of pixel value)

In the present invention, it is possible to select various configuration methods of the pixel value (one pixel value or both pixel values). More specifically, when the image is a plurality of images, the pixel value is It is possible to set it as a vector value which arranged the scalar value which the pixel of the same point which passed each of the said several image has arranged.

At this time, the plurality of pixels may be a plurality of still images constituting a moving image. In such a case, the present invention does not impair spatial resolution nor impair temporal resolution when obtaining the new pixel value. In addition, a preferable application example in such a case is, for example, the case where the moving image is a dynamic CT image acquired by a medical image diagnosis apparatus.

In the second method of constructing the pixel value, the image is a plurality of types of images relating to the same subject, and as described above, a scalar value of pixels having the same point passing through each of the plurality of types of images is arranged as a vector value. It is possible. Further, a preferred embodiment in such a case is, for example, an image obtained by a multi-window imaging method of a SPECT device or a PET device, each based on gamma rays generated from radioisotopes different from the plurality of types of images, respectively. The case is another image. Another preferred application example is the case where the image is a color image, and the images are different images obtained by decomposing into three primary colors of light, respectively.

In the third method of constructing the pixel value, the image is one image, and the pixel value may be configured as the one image. More specifically, for example, the pixel value can be configured as a vector value in which a scalar value of an arbitrary pixel on the one image and a scalar value of a pixel in the vicinity of the pixel are arranged. In addition, according to the configuration method of such pixel values, the present invention can be applied to a so-called digital image.

(Preferred apparatus configuration for realizing image processing of the present invention)

Finally, an apparatus configuration for realizing the above-described image processing method includes: fitness quantification means for determining a goodness of fit between a pixel value configured as a vector value or a scalar value and another pixel value separately configured for a pixel constituting an image; When a new pixel value for is constructed using the other pixel value, if the goodness of fit is large, the contribution of the other pixel is increased; if the goodness of fit is small, the contribution of the other pixel is made small, It is preferable to have pixel value calculating means constituting a new pixel value.

EMBODIMENT OF THE INVENTION Hereinafter, embodiment of this invention is described, referring drawings. In the following, first, a more general description (item No. “I” and the same below) regarding the outline of the “coherent filter” according to the present invention is given, and then the coherent filter is applied to various image processing. (Item numbers “II” and “III” to “Ⅷ”) will be described in order.

Subsequently, following the description of these various application examples, the most common form of the coherent filter of the present embodiment and its operation example (item number “Ⅸ”) will be described. The number “Ⅹ”) will be summarized and explained.

(I General Description of Coherent Filters According to the Present Invention)

First, the preliminary explanation for explaining the "coherent filter" which concerns on this invention is given.

(I-1 pixel value (v (x)))

In general, a digital image acquired through an imaging means such as a camera is composed of a plurality of pixels (or the image can be regarded as a set of such pixels). In the following description, the position of the pixel is represented as a vector x (that is, a vector of coordinate values), and the value (for example, a numerical value representing shades) of the pixel x is represented as a K-dimensional vector. In the case of a two-dimensional image, the pixel x is a two-dimensional vector indicating coordinate values (x, y) indicating a position on the image. “Pixel value v (x)” defined for an arbitrary pixel x,

[Equation 3]

Mark as.

V ₁ (x), v ₂ (x),... On the right side of Equation 3 above; , v _K (x) are hereinafter referred to as "scalar values" for the pixel x.

For example, when an image is a "color image", since each pixel has brightness (scalar values) of three primary colors (red, green, and blue), the pixel value (v (x)) of each of these pixels has the same dimension. It can be regarded as a vector of K = 3 (assuming that the subscripts are, for example, "red", "green", and "blue" in the right-hand terms of Equation 3, see Equation 3.1 to be described later.) For example, when a picture is a moving picture composed of k still pictures, and each pixel of the nth picture has a scalar value v _n (x), k still pictures and the same common point (coordinate) are the same. K-dimensional vector values (v _n (x) = (v ₁ (x), v ₂ (x), ..., v _K (x)) which are arranged by arranging pixel values (scalar values) of the pixel x Is a pixel value as a vector value described below.

(I-2 goodness of fit to risk (p (x, y)) and weight (w (p (x, y))))

Consider the appropriate set of pixels N (x) for the pixel x (the set N (x) may include the pixel x). Next, each of the elements of N (x) Consider a weight w (p (x, y)) between the pixel y and the pixel x. This weight w (p (x, y)) has the following properties.

(I-2-1 goodness of fit to risk (p (x, y)))

First, the meaning of the function p (x, y) that influences the value of w (p (x, y)) will be described. This p (x, y) is a means for quantifying the "fit" in the present invention, and generally speaking, how similar is the pixel (x) and the pixel (y∈N (x)) to what meaning (for example, The specific numerical value which shows the degree of the statistical difference recognized between the said pixel value v (x) and v (y) of both pixels x and y is given.

More specifically, for example, when p (x, y) gives a small value, the pixel x and the pixel y are “statistically significant” between the pixel values v (x) and v (y). It is judged that there is no difference (= great suitability), and it is likely to be similar, and when p (x, y) gives a large value, it is judged as "statistically significant difference (= good suitability)". will be.

By the way, the pixel values v (x), v (y) (or scalar values v ₁ (x),…, v _K (x) and v ₁ (y),…, v _K (y)) are necessarily noise. You must think that it contains. For example, in the case where an image is acquired by a CCD image pickup device, noises due to irregular variations in the dark current in the device or the amount of light incident from an external system exist for each pixel constituting the image.

Since such noise generally takes different values for all the pixels, even if the pixel x and the pixel y reflect the same object (for example, in the external world), they do not have the same value on the actually observed image. There is something that does not. In other words, in the case of the pixels x and y reflecting the same object, assuming a situation where the respective noises are removed, for example, they are displayed on the image as representing the same object (= recognized as such). And both have essentially the same (or very close) pixel value.

Therefore, based on the nature of the noise, the concept of "null hypothesis", which is well known as a statistical test for p (x, y), is used to describe this p (x, y). You can say In other words, the null hypothesis H "the pixel x and the pixel y have the same pixel value when the respective noises are removed" is in other words "v (x) = v (y). However, if the difference caused by the noise of both pixels is eliminated (that is, when such a proposition is established, it is considered that "the two pixels (x, y) are similar (= good suitability)"). The function (p (x, y)) is the risk (or significance level) when rejecting this hypothesis (H) (in this case, p (x, y)) is the area [0, 1]. It is defined as a function of (p (x, y) ∈ [0, 1]).

Therefore, when the risk (p (x, y)) is large, that is, when the risk of rejection is an error is high, it is highly likely to satisfy the hypothesis (H), and conversely, when the risk of rejection is an error is small. It can be said that the possibility of not satisfying the hypothesis (H) is high. (In addition, although it is well known in the statistical test, it does not mean that it is "true" even if the hypothesis (H) is not "dismissed.") The proposition indicated by the hypothesis (H) only means that it cannot be denied.)

 (I-2-2 weight (w (p (x, y))))

The weight w (p (x, y)) is a function of the above mentioned risk (p (x, y)) (more generally a function of goodness of fit, (x, y) can be configured to be w (ρ (x, y)), and each combination of x and y to obtain this weight (w (p (x, y))) The weight function (w) acting on the risk (p (x, y)) obtained for s generally has a function of embodying the above "rejection." Specifically, the risk (p (x, y)) is large. In this case, the value of the weight function (w), that is, the weight (w (p (x, y))) takes a large positive value, and vice versa, a small positive value (or “0”). (The specific form of the weighting function w will be described later.) That is, the weights w (p (x, y)) correspond to the pixel (x) and the pixel (y) according to the hypothesis (H). If the stated propositions are satisfied, they take a large value, and vice versa. In this case, a small value is taken, and as an example, it may be configured such that only two values of w can be "0" or "non-zero".

In addition, the relationship between the hypothesis H, the risk ratios p (x, y), and the weights w (p (x, y)) described above is collectively shown in FIG. The weight function w (t) is more generally referred to as a non-negative monotonic increasing function defined by t 함 [0, 1], and the corresponding w (t) is satisfied. The property to be done should be at least like that.

(I-3 coherent filter)

By the above preliminary explanation, the "coherent filter" according to the present invention is derived as follows. In other words, the above-described weights w (p (x, y)) are calculated for all the pixels y that are elements of the set N (x) for any pixel x constituting the image. Subsequently, using the plurality of weights w (p (x, y)), a new scalar value v ′ _k (x) constituting the pixel x is calculated by the following equation. In other words,

[Equation 4]

Provided that k = 1, 2,... , K. Then, the pixel value (new pixel value) (v '(x)) after the conversion of the pixel x is obtained using (v' _k (x)) obtained by the above expression (4).

[Equation 5]

Configure as.

Accordingly, the pixel value (v (y) = (v ₁ (y), v ₂ (y), ..., v _K (y)) (including the case where y = x)) represented by Equation 4 above is represented by v. The format of this embodiment of the "coherent filter" according to the present invention is a filter for converting to '(x) = (v' ₁ (x), v ' ₂ (x), ..., v' _k (x)). to be. This indicates the weighted average value of the scalar values v _k (y) constituting the pixel value, as apparent from this marker.

This treatment results in the following results. That is, the pixel value v '(x) takes the same pixel value apart from the pixel x into the noise (which is highly likely to satisfy the proposition of the hypothesis H). It represents a vector composed of weighted average values v ′ _k (x). If a sufficient number of such pixels y exists, the pixel value v '(x) does not deviate from its true value that the pixel x should originally have, and the value obtained by suppressing only noise by the above-described averaging operation. Will have

In addition, even when the risk (p (x, y)) is small, and thus the null hypothesis H is "dismissed" and the weight (w (p (x, y))) becomes small, As can be understood from the mark, it is not necessarily limited to completely "dismissed". This depends on the specific form of the weight function w described below, but even when the risk p (x, y) is close to “0” (= 0%), w (p (x, y)) ≠ 0 (However, it is a smaller positive value than when p (x, y) is close to " 1 "). ) = v (y). That is, it is not a complete rejection, but a small contribution is acceptable. (In this case, if w (p (x, y)) = 0, the (Equation 14 to be described later).

Such a process can generally be said as follows. That is, when there are (plural) pixels x constituting an arbitrary image, the suitability of the pixel x with any arbitrary pixel y (in the above case, y∈N (x)) is quantified. (Based on p (x, y) in the above), and when the goodness of fit is large, in the weighted averaging process using the pixel value v (y), a large contribution to the pixel y is acknowledged. In the case where the goodness of fit is small, only a small contribution is to be accepted, which can be said to be an image processing method that effectively suppresses the noise of the pixel x. In other words, when the pixels x and y are "similar", the pixel y is more contributing to the averaging process. When the pixels x and y are "similar," the pixels y are mostly or completely ignored. You can say that again.

By performing such a process on the entire image, a very high noise suppression effect can be exhibited with little blurring of the image. Moreover, it is not limited to the use of noise suppression, For example, in the field of pattern recognition, the outstanding effect can be exhibited by making a weight function or a coherent filter into a preferable specific form. Incidentally, the specific effects of the above-described effects will be described below in the description of various image processing.

(II Various Application Examples: Classification by Difference in Construction Method of Pixel Value v (x))

The present invention can be applied to various fields based on the general form of the coherent filter described above. And this can be summed up as shown in FIG. 2 as an example. In addition, in FIG. 2, various application examples in various fields are shown classified according to the difference in the construction method of the pixel value v (x). That is, according to this Fig. 2, the pixel value v (x), which is the above-described vector value, is " multiple images ", " multiple types of images " It can be seen that it is possible to configure by each of the methods, and it is understood that various application examples according to the present embodiment are classified into each of these.

Hereinafter, according to the classification table shown in FIG. 2, the three types of pixel values v (x) shown in FIG. The method of concretely realizing the construction method will be described in turn (however, unlike the procedure shown in FIG. 2).

(When the present invention is applied to III X-ray CT imaging 1; Dynamic CT imaging (removes noise from multiple images))

First, the case where the coherent filter is applied to so-called "dynamic CT" imaging of an X-ray CT apparatus will be described. Here, the X-ray CT apparatus 100 is an X-ray tube 101, an X-ray detector 102, a data collection unit 103 (DAS; Data Aquisition System), and a preprocessor 104 as shown in FIG. The memory unit 105, the reconstructing unit 106 and the image display unit 107, the control unit 108 for controlling each unit, and the input unit 109 capable of various settings and inputs such as X-ray irradiation conditions, photographing modes, and the like. Is done.

According to this X-ray CT apparatus 100, the object P is irradiated by radiating the X-ray onto the object P while rotating the X-ray tube 101 and the X-ray detector 102 around the object P. The inside of the can be observed as other CT images such as tomographic images.

At this time, each component acts as follows schematically. That is, the X-rays that have passed through the X-ray tube 101 and pass through the object P by applying the high voltage of the high voltage device 101a are converted into analog electric signals by the X-ray detector 102, hereinafter, the data collection unit. The digital conversion processing by 103, various correction processing by the preprocessing unit 104, and the like are stored in the memory unit 105 as projection data. The tomographic image of the X-ray tube 101 and the X-ray detector 102 is smaller than that of the reconstructing unit 106 based on projection data obtained as a result of one round (= 360 °), for example, around the object P. Other CT images are reconstructed, and the CT images are displayed on the image display unit 107. The reconstructed CT image can also be stored in the storage device 10M.

The "dynamic CT" imaging described above means that the X-ray tube 101 and the X-ray detector 102 repeatedly photograph the same portion of the subject P. (Repeat scan. Shooting is often performed.) The photographing method is obtained by continuously obtaining projection data and continuously reconstructing the processing based on the projection data to obtain a series of time series images (in this case, the image display unit 107). The image display is controlled to be carried out after a certain time, for example, at the start or end point of the scan in accordance with the collection of projection data on which the image is based, for example, by a counter (not shown).

Therefore, the image acquired and displayed in this way becomes a so-called moving picture composed of a plurality of time-series still images like a movie or the like. In addition, such an imaging method is typically used to inject a contrast agent into the subject P, observe and analyze the change over time, and analyze conditions of other lesions, such as narrowing or occlusion of blood vessels. . Moreover, the method of performing CT imaging of the same site only twice before and after administration of contrast medium can be considered to be a broad dynamic CT imaging.

In the conventional "dynamic CT" imaging, for example, a change (for example, a change in the concentration of the contrast agent or a change in breathing, etc.) is generally considered in the subject P during the imaging of K times. ), In order to suppress image noise without compromising the spatial resolution, smoothing in the temporal direction is inevitable. As a result, the damage which the time resolution impairs was unavoidable.

However, since the image acquired by dynamic CT imaging is a moving image as described above and is performed for the purpose of observing the temporal change in detail, it is not inherently preferable that the time resolution is impaired.

When the coherent filter according to the present invention is used, the following processing can be suppressed for all k still images (multiple images) without impairing the time resolution. Filter processing ”).

First, as described above, the pixel x defined for the k still images, which are the moving images obtained as described above, is defined as the pixel value v (x).

(Equation 3)

Can be configured. Where subscripts 1, 2,... , K is a number assigned through each of k k still pictures (refer to the description regarding Equation 3 above).

Subsequently, the specific form of the weight function w1 in this case is given by the following equation (6).

[Equation 6]

However, y ∈ N (x), and this set N (x) may be arbitrarily set for the pixel x (= may be set according to any criteria). However, in reality, the pixel x and the pixel y at a position far from the pixel x are hypothesized, "v (x) = v (y)". However, since the possibility of satisfying the removal of the difference due to noise of both pixels is generally low, limiting the set N (x) to the reference that the set is a set of pixels close to x is such as to improve operation speed. There is practical significance.

Therefore, here, as an example, set N (x) is a set of pixels contained in the rectangular area | region around it centering on the said pixel x. More specifically, as a set N (x), for example, when all the pixels constituting one still picture, which is now paying attention, are 128x128 pixels, a portion of 3x3 pixels centering on the pixel x is provided. In the case of 512 x 512 pixels, an area of 13 x 13 pixels centered on the pixel x may be used.

Σ _k in Equation 6 is a standard deviation of noise estimated on the assumption that each pixel of the _k- th still image has a certain degree common to all of them, while C is a weight w1 (p (x, y)) is a parameter that makes it possible to determine and control the degree of the action in the case of substitution in Equation (4).

Hereinafter, the above descriptions for σ _k and C will be given.

First, σ _k in the above expression (6) will be described (hereinafter, it will be described as variance σ _k ² ). As described above, σ _k ² is the dispersion of the noise component of the scalar value of each pixel on the k-th still image. The variance σ _k ² of the above equation (6) is assumed to include noise having a variance σ _k ² which is a constant value with respect to the scalar value of each pixel of the k-th image. In general, such assumptions have sufficient justification against the backdrop of the following.

First, the CT image noise is constant when the size of the subject P, the structures of the X-ray tube 101, the X-ray detector 102, the reconstructing unit 106, and the like are constant, and the energy of the irradiation X-ray is constant. Is determined by the product of the irradiation current of the X-ray tube, that is, the X-ray tube 101 in proportional relationship with the irradiation time (so-called tube current time product (mA · s)). On the other hand, noise in CT images is additive, and it is also known that the noise of the CT image is generally large. That is, the true value (contribution of noise) to any scalar value (v _n (x) (n = 1, 2, ..., K)) constituting the pixel value v (x) of the arbitrary pixel x. If the value of) is set to v _n ⁰ (x), the difference value (v _n (x) -v _n ⁰ (x)) is approximately equal to the Gaussian distribution of 0 and variance (σ _k ² ). Irradiation X-ray dose to tube current time product (m · As) and noise dispersion (σ _k ² ) are approximately inversely related.).

The dispersion σ _k ² also depends on the position of the pixel x itself (for example, each coordinate value x == (x, y) as described above), but a normal X-ray CT apparatus ( In 100, a so-called "wedge" or "X-ray filter, which is constituted by a physical X-ray filter (for example, copper foil or metal mass, etc.) for controlling the X-ray irradiation amount between the X-ray tube 101 and the X-ray detector 102. ). (Not shown), so that this can be ignored. This is because the wedge adjusts a part or all of the X-ray tube irradiated so that the same amount of X-rays can be detected in any X-ray detector 102 by using a substance P having a density almost equal to that of water. This is because it has an effect of absorbing or shielding, and accordingly, such a wedge effectively produces the effect of making the dispersion of noise σ _k ² a constant value almost independent of the position of the pixel x (i.e., This wedge is generally provided for the purpose of effectively utilizing the dynamic range of the X-ray detector 102.).

As described above, it is reasonable to estimate the variance σ _k ² , which is a constant value, for all the pixels of the k-th still image obtained by k-CT still images obtained by dynamic CT imaging. Of course, it will be possible to easily extend the present embodiment for the case where the dispersion is different for each pixel (described in detail in section X-3 to be described later).

Subsequently, in order to specifically calculate the above expression (5), it is a matter of which numerical value is assigned as the variance σ _k ² . Such a problem is usually caused by the fact that although the shape of the noise distribution can be assumed (the Gaussian distribution in the above), the specific value of the dispersion (σ _k ² ) is often unclear.

In general, the imaging may be performed by changing the irradiation dose (X-ray tube current x irradiation time (mAs)) for each shooting.

Then, in the k-th image (k = 1, 2, ..., K), the dispersion of noise of the scalar value of each pixel is set to σ _k ² , and the dose of radiation used to capture the k-th image is set to R _k . Σ _k ² is proportional to R _k . Therefore, if σk ₀ ² can be specified for at least one k = k ₀ ,

Σ _k ² can be estimated accurately.

(Also this situation is suitable) in this embodiment, it is possible to perform estimation of the specific values of σ _k ² in the following manner with respect to at least one of k.

Expected value (E [σ) for dispersion (σ _k ² ) by actual measurement using N times (1 <N ≦ K) images that can assume little change in the subject P during K shots. _k ² ]) is valid. For the sake of simplicity, the radiation doses of these N images are the same, so k = 1, 2,... Regarding N, σ _k ² is assumed to be constant (write σ ² ). Each scalar value (v ₁ (X _f ), v ₂ (X _f ), ..., v _K (X _f constituting the pixel value v (X _f ) of any pixel X _f of these N pixels. The noise included by)) is expected to be according to the Gaussian distribution of mean 0 and variance (σ ² ) as described above.

The average of these,

[Equation 7]

Using, the expected value (E [σ ² ]) for the true variance (σ ² ),

[Equation 8]

It can be obtained as

The expected value E [σ ² ] of the variance can be considered as valid for all the pixels x of all k still images as described above, and is used as a substitute for the true variance σ ² . The value is guaranteed to be more than certain. Therefore, in the actual operation of Equation 6, E [σ ² ] may be substituted into σ ² of Equation 6.

More specifically, such E [σ ² ] may be determined based on actual values based on, for example, the first and second still images of k still images (let's say the above equations (7) and (8). Equivalent to distant N = 2). In addition, for the pixels X _f provided for the actual calculations of the equations (7) and (8), for example, only the appropriate pixel (X _f ) excluding the portion where the air or bone is being imaged is selected (multiple selections). The average of all the obtained E [σ ² ]) may be taken. In addition, in general, the present invention may be devised to suppress photographing caused by the movement of the subject P. In addition, the evaluation of the variance of noise (σ ² ) in Equation 6 is described later. Supplemental Embodiments].

The irradiation dose estimated by that the σ _k ² is proportional to R _k σ _k ² correctly, even if it is not constant in the image recording sheets of these N will be able to dance to facilitate.

Subsequently, the parameter C of the above expression (6) will be described. First, in Equation 6, the thinking method of the risk ratio p (x, y) described in the general form is included as follows. In other words, the form in the root of the right-side molecule of Equation 6 corresponds to the corresponding χ ² value that follows the so-called χ square distribution, divides this by (2 σ) ² , and puts the entire parenthesis on the shoulder of e. The risk (p1 (x, y)) is itself. In other words,

[Equation 9]

to be.

Equation 6 relates to p1 (x, y) represented by Equation 9,

[Equation 10]

It must have been.

 In addition, A is an integer and is normalized so that p1 may be a value of (0-1).

As a result, in Equation 6, the risk (p (x, y)) described in the above general form is not shown in the positive, but the weight (w1 (p (x, y))) state is clearly as described above. It can be regarded as a function of the risk (= p1 (x, y)) (Equation 10), that is, a function of fitness (However, the risk and the goodness-of-fit increase as one increases as the other increases. Relationship).

As shown in Equation 10, the parameter C has an effect of determining how sensitive the weight w1 (p (x, y)) is to the risk p1 (x, y). There is. In other words, when C is made large, w1 (p (x, y)) is close to 0 only by slightly decreasing p1 (x, y). Moreover, when C is made small, such a sensitive reaction can be suppressed. Moreover, as C, it may be made specifically about 1-10, Most preferably, it is good to set it as C = 3.

In this embodiment, the similarity judgment regarding both pixels (x, y), in other words, the determination of rejection of the null hypothesis (H) above with respect to both pixels (x, y) is as described above. Based on (p1 (x, y)), it is determined by what is called χ-square test (statistical test).

In addition, as can be seen from the expression of Equation 6, in the present invention, the risk (p (x, y)) is calculated for each combination of x and y, and then the weight (w (p (x, y)) It is not necessary to go through the order of obtaining), and it is possible to have a configuration in which (wop) as a synthetic function is calculated directly without specifically calculating the risk (p (x, y)).

As described above, the variance σ ² is estimated (for example, E [σ ² ] in Equation 8), and the parameter C is appropriately obtained (for example, C = 3). All pixels included in a set N (x) defined for an arbitrary pixel x (for example, 3x3 pixel area centered on the pixel x as described above) using Specific weight (w1 (p (x, y))) can be obtained for (y). Subsequently, specific numerical calculation of the coherent filter can be performed by using w1 (p (x, y)) in place of w (p (x, y)) in Equation (4). As a result, the pixel values v '(x) = (v' ₁ (x), v ' ₂ (x), ..., v' _K (strongly suppressed noise without impairing the time resolution as well as the spatial resolution) x))) (= Equation 5), i.e. k such still images or moving images can be obtained.

4 is a diagram illustrating such image processing so as to be conceptually understood. That is, first, in Fig. 4A, 1, 2,... In the k still images, a rectangular region (N ^{3 × 3} (x)) of 3 x 3 pixels centered on the pixel x is assumed for an arbitrary pixel x. If the pixel at each left corner of the rectilinear region N ^{3 × 3} (x) is y ₁ , this pixel y ₁ has the pixel value v (y ₁ ) as shown in FIG. 4 together. have.

And, of the pixel value (v (y ₁₎₎ scalar values making up the _{(v 1 (y 1),} v 2 (y 1), ... v K (y 1)) and the pixel value (v (x)) By each of the scalar values v ₁ (x), v ₂ (x), ..., v _K (x), the weight w1 (p (x, y ₁ )) is calculated by the above equation (6) ( (B) of FIG. 4). The same applies to the remaining pixels y ₂ ,..., Y _{8 of the} rectangular region N ^{3 × 3} (x), and finally w1 (p (x, y ₁ ) as shown in FIG. )),… , w1 (p (x, y ₈ )) and w1 (p (x, x)) are obtained (in this case, the risk factor (p (x, x)) is &quot; 1 &quot; The weight w1 (p (x, x)) is also "1" from equation (10) (= maximum weighting).

Subsequently, the weights w1 (p (x, y ₁ )), ..., w1 (p (x, y ₈ )), w1 (p (x, x)) obtained in this manner are k-th corresponding pixels. Multiply each by a scalar value (v _k (y ₁ ), v _k (y ₂ ), ... v _k (y ₈ ), v _k (x) of the pixel and take the sum (corresponding to the numerator of Equation 4 above), Dividing this by the sum of the weights w1 (similar to the denominator of Equation 4) with respect to the rectangular region N ^{3 × 3} (x) reduces the noise of the pixel x of the kth image. The scalar value v ′ _k (x) can be obtained (FIG. 4C). K = 1, 2,... Noise is suppressed using the same weights w1 (p (x, y ₁ )), ..., w1 (p (x, y ₈ )) w1 (p (x, x)) for all pixels of K Pixel value ( _v'k (x) = v ' ₁ (x), v' ₂ (x), ..., v where noise of pixel x is suppressed by obtaining scalar value _v'k (x)) _K (x))) is obtained. Repeating the above operation for all the pixels x yields k images with noise suppressed.

The image composed of the pixel value v '(x) calculated in this way is as shown in Figs. Fig. 5 shows the original image (k = 11) before applying the coherent filter among all K = 23 still images, and Figs. 6 and 7 show the case where the coherent filter is applied for k = 11 and 23. Each image is shown. It can be seen that the random noise shown in the original image of FIG. 5 is sufficiently suppressed in FIGS. 6 and 7.

In addition, each process described above may be implemented according to the flowchart shown in FIG. 8, for example, and an example is performed in order to implement calculation, image display, etc. according to each process on the actual X-ray CT apparatus 100, for example. For example, as shown in FIG. 3, the image processing unit 110 constituted by the dispersion value estimating unit 111, the weight calculating unit 112, and the pixel value calculating unit 113 may be provided.

Among these, the weight calculator 112 obtains the weight w1 (p (x, y)) directly by the pixel values v (x) and v (y) in the above-described order. Therefore, the calculation unit 112 does not specifically calculate the value of the risk ratio p1 (x, y) (i.e., a device for calculating the weight immediately after the "risk calculation unit (incorporating the" fit quantification unit "referred to in the present invention)). In addition to the above-described configuration, the "risk calculation unit (compatibility quantification unit)" which calculates the value of the risk ratio p1 (x, y) specifically, and the weight (w1 (p (x, y))) based on the output thereof. The weight calculation unit 112 may be configured to take the order of two stages called the “weighting operation unit”, which is obtained by any means, and the weight calculation unit 112 may include the variance σ ² estimated by the variance value estimating unit 111 and v (x) and v (y). ), The weight w1 (p (x, y)) is calculated.

In addition, the pixel value calculating section 113 uses the pixel values v (x), v (y) and the weight values (w1 (p (x, y)) numerically calculated by the weight calculating section 112 to determine the pixel values. Calculate (v '(x)) That is, the calculating part 113 actually performs the process of suppressing the noise of the underlying image, that is, applying the coherent filter (hereinafter referred to as "coherent filter is applied").

In the dynamic coherent filter process described above, when a coherent filter is applied to a moving picture composed of k still images, the processing of the image processing unit 110 once reconstructs all still images, Although it may accumulate in the memory | storage device 10M, and may apply a coherent filter to these later as a post-processing, this invention is not limited to this form, The above-mentioned continuous scan, continuous projection data collection, continuous reconstruction, In the flow of continuous display, the process of applying the coherent filter in real time may be performed (hereinafter, these are referred to as "real time coherent filter process").

In a preferred embodiment of the real-time coherent filter process, each time a new image is photographed and reconstructed, the following process is performed. Image numbers M, M-1,... From the first obtained image (image number 1) to the latest image (image number M). And k-dimensional vector values (v (x) = (v _M ) by arranging the pixel values (scalar values) of the pixels x having the same common point (coordinate) on k still images having M-K + 1. (x), v _{M -1} (x), ..., v _{M -K + 1} (x))). In this way, the coherent filter can be attached in the same manner as in the above "dynamic coherent filter process". However, the pixel value calculating section 113 does not actually calculate all the elements of the pixel value v '(x), but rather a scalar value v _M ' (x) corresponding to the latest image (image number M). Calculate only. As a result, the calculation speed is improved, so that the latest image in which noise is suppressed in real time can be displayed.

As another preferred embodiment of this &quot; real-time coherent filter process &quot;, a coherent filter is attached in the same manner as above when the first k images are shown, and v ₁ '(x),... , v _K ′ (x), and then, k ( _k ) vector values are obtained using k stationary images having image numbers (M, M-1, ..., M-K + 1). _M (x), v _{M -1} '(x), ..., v _{M -K + 1} ' (x)), and may be configured to apply the real-time coherent filter process. In addition, it is convenient to configure so that the dimension K of the pixel value vector v (x) can be changed at any time by manual setting or automatic setting during these real-time coherent filter processes.

(In the case of applying the present invention to IV X-ray CT imaging; 2 (to obtain one pixel with reduced noise))

Subsequently, a specific example of obtaining one image, which is different from the above-mentioned dynamic CT and reduced noise, will be described. Conventionally, if one wants to obtain one CT image with generally low noise, in order to achieve this, a simple average in time direction (for example, k CT images taken K times using a normal X-ray tube current) is obtained. Using the concept of the pixel value v (x) in the present embodiment, the simple average is {v ₁ (x) +… + v _K (x)} / K). Two methods can be considered, a method of obtaining and a method of obtaining one image by one imaging using a normal K-time X-ray tube current. Either way, the dispersion of noise is not changed to be 1 / K times of one image obtained by photographing once using a normal X-ray tube current. As a result, radiation X-ray dose and noise dispersion are inversely related). Therefore, conventionally, if the performance of the X-ray tube is allowed because of the efficiency of the operation, the latter one is photographed at once, and if the X-ray tube cannot withstand a large amount of X-ray tube current, the former is repeatedly photographed K times. It was generally carried out to obtain one CT image by the method.

However, in these systems, the amount of irradiation X-rays increases as the noise is to be suppressed. That is, the exposure amount of the subject P increases.

Here, by using the coherent filter according to the present invention, it is possible to suppress the noise included in one such CT image without increasing the exposure amount to the subject P.

First, in the present embodiment, k CT images are taken at a lower X-ray dose than usual during the time when it can be assumed that there is no change in the subject P. In this case, therefore, with respect to the pixel x defined for k CT images as in the dynamic coherent filter process,

(Equation 3)

The pixel value v (x) can be configured as if. In this case, since the X-ray irradiation amount is small, the dispersion σ ^{2 of} noise included in each CT image becomes relatively large.

Subsequently, as the weight function in this case, the same concrete form as in Equation 6 is given.

In this case, however, since the final image to be obtained is one CT image, the following method can be considered.

For example, in the same way as in the above-mentioned dynamic CT photography, a coherent filter is attached to all k images, and the resultant k images are weighted averaged or simply averaged to form a single CT image of interest. The coherent filter may be applied to only one of the k images selected arbitrarily (the calculation in FIG. 4C is performed only for the selected one image).

That is, according to the former method, a calculation for obtaining one image obtained by averaging these images from k images resulting from applying a coherent filter to all k images is specifically, for example,

[Equation 11]

to be.

V ^{* '} (x) on the left side of the above expression (11) is the pixel value (scalar) of the pixel x of one image obtained by a simple average result. At this time, the variance (σ ² ) ^* of noise included in V ^{* ′} (x) is approximately,

[Equation 12]

Become,

In addition, since w (p (x, y)) ≥ 0, the term summed at the right side of the expression (12) is always smaller than one. Therefore, the noise suppression effect according to the present embodiment is superior to the noise suppression effect (σ ² / K) achieved by the conventional method. Conversely, if the noise suppression effect equivalent to the conventional method is to be obtained by using the present embodiment, the required X-ray irradiation amount is solved less than that of the conventional method, and therefore, the exposure amount of the subject P can be reduced.

Effects obtained using the present embodiment are as shown in Figs. 9 to 11, for example. Fig. 9 shows an image obtained by simply averaging in a conventional manner, i.e., images repeatedly taken K times in the time direction, and Fig. 10 is obtained by attaching a coherent filter to all k images according to the present embodiment, The image comprised by the simple average of k images is shown. It can be seen that the random noise shown in Fig. 9 is sufficiently suppressed in Fig. 10. (In addition, in Fig. 9 to Fig. 11, when the 0.5 second scan is taken three times (K = 3), the above-mentioned set ( N (x)) is an image under a condition of C = 3 using a rectangular region of 13x13 pixels.)

In addition, since FIG. 11 is very excellent in the noise suppression effect of FIG. 10, it is comprised in consideration that it becomes an unfamiliar image compared with the conventional image, Specifically, it is the image which took the average of FIG. 9 and FIG. That is, in the image of FIG. 11, the pixel value is represented by v ** ′ (x), and the pixel value of FIG. 9 is represented by v (x) (= {v ₁ (x) +… + v _K (x)} / K). When you do,

[Equation 13]

In this case, u is 1/2. In addition, in this invention, it is not limited to this embodiment itself (FIG. 10 is obtained), For example, the value of u of Formula (13) (where 0 <= u <= 1) can be adjusted according to a patient's preference. The configuration so that it is desirable is also preferable.

(3 when applying the present invention to V X-ray CT imaging; multi-energy imaging (reduces noise of plural kinds of images))

Then, the case where this invention is applied to what is called "multi-energy imaging" is demonstrated. Multi-energy imaging means the energy of X-rays to be irradiated by changing the voltage applied to the X-ray tube 101 or inserting an X-ray filter composed of a thin metal plate or the like between the X-ray tube 101 and the object P. The method of imaging the same subject P by the several energy distribution obtained by changing a distribution in various ways and saying it is said. In this case, although the subject P does not change, the image changes depending on the energy distribution of the X-ray, and as a result, a plurality of types of images are obtained.

Here, if K images are taken while changing the energy distribution, when the scalar values of the pixels x of the K kind of images obtained by the K images are listed, the pixel values (v (x) = (v ₁ (x), v) ₂ (x), ..., v _K (x))). Therefore, the dynamic coherent filter processing of this pixel value v (x) (1 when the present invention is applied to III X-ray CT imaging; dynamic CT imaging (removing noise from multiple images)) and By applying the same means, a coherent filter can be attached, and as a result, K kinds of images in which random noise is suppressed can be obtained.

(When the present invention is applied to medical imaging apparatus other than VI X-ray CT apparatus)

Hereinafter, the coherent filter according to the present invention is a medical imaging apparatus other than an X-ray CT apparatus, for example, the magnetic resonance imaging apparatus (so-called "MRI apparatus"), a nuclear medical diagnostic apparatus ("synthesis camera, SPECT apparatus"). Or PET device)), an embodiment applied to an X-ray see-through device and the like will be described.

(When the present invention is applied to VI-1 MRI device)

First, the application example of this invention regarding MRI apparatus is demonstrated. The present invention relates to a photographing method, which is conventionally referred to as "averaging", "high speed photographing", "three-dimensional photographing", and "photographing method using a plurality of pulse sequences". It is possible to apply for.

First, averaging is, as its name suggests, to reduce image noise by averaging, and repeatedly photographing while no change occurs in the subject P, thereby simplifying the obtained multiple (k) images. This is a photographing method that obtains one image on average. By arranging the scalar values of the pixels x of the k images, the pixel values v (x) = (v ₁ (x), v ₂ (x), ..., v _K (x)) can be achieved. . The weight function w may also be represented by, for example, Expression (5). Therefore, the dynamic coherent filter processing of this pixel value v (x) (1 when the present invention is applied to III X-ray CT imaging; dynamic CT imaging (removing noise from a plurality of images)) and A coherent filter can be attached by applying the exact same means, and as a result, a plurality of images obtained by suppressing noise superior to that by averaging are obtained. Alternatively, one image obtained by suppressing noise may be obtained by applying the same processing as in (In case of applying the present invention to IV X-ray CT imaging, 2; (one image obtained with reduced noise)).

In addition, high speed photography is a method of obtaining a moving picture showing a change over time by repeatedly photographing a change in the subject P, which is also called dynamic MIR. In this case, there is no change in obtaining a plurality of (k) still images, as in the case of averaging (1 when the present invention is applied to III X-ray CT imaging; dynamic CT imaging (from multiple images). Noise can be suppressed by applying a coherent filter in the same manner as the dynamic coherent filter process of removing noise). In high-speed imaging, the same processing as in the real-time coherent filter process of (III. Application of the present invention to X-ray CT imaging: Dynamic CT imaging (removing noise from multiple images)) may be applied. good.

Three-dimensional photography is a method of photographing three-dimensional images at once. In this imaging method, averaging (i.e., a process of repeatedly photographing for the purpose of suppressing noise and taking a simple average) is performed, or high-speed photographing is repeatedly performed when a change occurs in the subject P. Photographing), and in these cases, the coherent of the above-described averaging and high-speed photography, except that the pixel x is represented as a three-dimensional vector having x = (x, y, z). It is exactly the same as the filter application method. In addition, in order to apply a coherent filter when 3-dimensional imaging is performed only once, the method demonstrated below (composing the pixel value v (x) from 1 image) can be applied.

On the other hand, in an MRI apparatus, another physical parameter can be made into an image by changing a pulse sequence. In the pulse sequence, a wide variety of images are obtained by changing the RF pulse and the gradient magnetic field in the application sequence, the intensity and time to be applied, and the pause time. It is known to name a few of the typically used pulse sequences, for example T2 * -weighted images, T1-weighted images, proton-density images, etc., by naming the physical parameters that most contribute to the contrast of the image. . In any case, although the subject does not change depending on the pulse sequence, another image is obtained, and as a result, a plurality of types of images are obtained substantially. Therefore, the coherent filter can be attached in exactly the same manner as in the case where the present invention is applied to V X-ray CT imaging; the noise of plural kinds of images is reduced.

More specifically, when K pictures are taken while changing the pulse sequence, the scalar values of the pixels x of the image obtained by the K pictures are arranged, and the vector (v (x) = (v ₁ (x), v) ₂ (x), ..., v _K (x))). Therefore, also in this case, the noise can be suppressed by applying the coherent filter by applying the same means as in the dynamic coherent filter process as in the case of the above averaging.

(In case of applying the present invention to VI-2 cineration camera, SPECT device or PET device)

Second, an application example of the present invention to a chinchion camera, a SPECT device, or a PET device will be described. These devices distribute a radiopharmaceutical in a subject by administering a drug (radiopharmaceutical) containing a radioisotope (hereinafter referred to as "RI") in a subject and detecting a gamma ray caused by spontaneous collapse of RI. It is used for the purpose of obtaining what is called a fan cushion map which shows the degree of the biochemical metabolism and the blood circulation, especially when a living body is used as a subject. The scintillation camera photographs the distribution of the radiopharmaceuticals as a two-dimensional image as if through a subject. In addition, the SPECT device and the PET device capture a three-dimensional image representing a three-dimensional distribution of the radiopharmaceutical by using a computed tomography technique.

Since the amount of gamma radiation normally detected in a scintillation camera, a SPECT device, or a PET device is very small, so-called "photon noise" due to the variation in the number of photons caused by the quantum mechanical properties of RI appears in the image.

Conventionally, in order to suppress photon noise without reducing the resolution of an image, there is no method other than increasing the detected gamma dose, and specifically, there has been a method of increasing the amount of RI administered and a method of lengthening an imaging time.

In cases where the amount of RI that can be administered can be increased (eg, dose escalation is allowed for laboratory animals for medical research), this can reduce the photon noise, but this approach can It is accompanied by an increase in radiation dose. In addition, in the case of diagnosing the human body (= a human being is a subject), such an increase in the amount of RI as described above is not freely allowed (legally regulated). In addition, it is impossible to increase the amount when imaging is performed by using radiation from which RI is contained in natural products.

In addition, when using the method of lengthening the shooting time, it is necessary that the subject does not move for a very long time (for example, several hours) in order to obtain an effective noise reduction effect. This condition is difficult to realize when a living body (particularly a human body) is used as a subject, and there is a problem that the processing capability (or cost performance) of the imaging apparatus is lowered due to a long imaging time. Therefore, in reality, it is inevitable to set an appropriate shooting time limiting the allowable time (for example, within 1 hour), and the appearance of strong noise in the image is unavoidable. Therefore, in order to suppress this noise, it is inevitable to perform it by smoothing an image, and therefore, it was considered unavoidable that the resolution of an image is impaired as a bad effect of smoothing.

In such a case, according to the present invention, the following processing can be performed. In other words, a shorter time of imaging is performed a plurality of times during a time equivalent to the appropriate shooting time, thereby obtaining a plurality of images with relatively high noise. Then, the same process as in (In case of applying the present invention to IV X-ray CT imaging; 2; (one image with reduced noise) is applied) is applied to produce an image in which noise is more suppressed without compromising spatial resolution. This can be achieved and can be achieved without increasing the amount of RI.

In addition, a photographing method that tracks changes over time in the distribution of radiopharmaceuticals in a subject by repeatedly photographing the same subject in a cineration camera, a SPECT apparatus, or a PET apparatus, such as an X-ray CT. Or "dynamic PET". In this case, too, the same method as the dynamic coherent filter processing or the real-time coherent filter processing for the case where the present invention is applied to III X-ray CT imaging is performed. The processing can be applied, and the noise of the image can be effectively suppressed.

In the PET or SPECT device, a plurality of types of RI are simultaneously administered to a subject by using RI emitting gamma rays of intrinsic energy for the type, thereby obtaining respective distribution images for a plurality of types of RI at once. So-called "multi-window photography" may be used. This is based on a technique for measuring the energy of the detected gamma rays and determining whether the gamma rays emitted by the plurality of RIs are determined according to whether or not the values correspond to a numerical range (energy window) set artificially. Doing.

More specifically, by setting an appropriate number of energy windows (for example, windows W1 and W2), an image based on gamma rays corresponding to one of the windows (for example, window W1), and another The above-described plurality of RI distributed images are obtained by one shot by forming images separately, such as images based on gamma rays corresponding to windows (for example, window W2).

That is, the same object is photographed using an imaging device to obtain multiple kinds of images under different processing conditions. More generally, it can be said that "a photographing method of obtaining different K kinds of images by one photographing using K kinds of RIs". Therefore, the coherent filter can be attached in exactly the same manner as in the case where the present invention is applied to V X-ray CT imaging; the noise of plural kinds of images is reduced.

That is, even in this case, the pixel values v (x) = (v ₁ (x), v ₂ (x), ... v _K (x)) are configured for the pixels x on such K kinds of images. can do. In addition, although the dispersion of noise included in each pixel value v (x) varies depending on the position of the pixel x and what kind of image it is, it is considered that it is almost proportional to the total amount of gamma rays detected for each type of image. It is good and because of this property, the dispersion of noise for each pixel x can be estimated. As described above, the present invention can also be applied to "multi-window photography" to which a coherent filter can be attached. As a result, K kinds of images in which random noise is suppressed can be obtained.

Nuclear medical diagnostic apparatuses include reconstruction methods (such as MU-EN reconstruction for fan beams and super-resolution reconstruction methods for fan beams and pararell Hall corimators, etc.) that are effective in improving the resolution, but these and the coherent filters according to the present invention. There is no principle restriction at all by combining the results, and as a result, it is also possible to create an image in which high resolution and random noise are sufficiently suppressed.

(In case of applying the present invention to VI-3 X-ray see-through device)

Third, an application example of the present invention to an X-ray see-through device will be described. First of all, the term "X-ray fluoroscopy device" used herein refers to a device that repeatedly photographs a predetermined amount of time (e.g., 1/24 second) while continuously emitting a low dose of X-rays to a subject. Frame ”, and by continuously displaying the X-ray image obtained for every frame, the apparatus which can observe as a so-called perspective image (movie image) is assumed. Therefore, in general, a device having a configuration called so-called "C arm" to "Ω arm" corresponds to the above, and the above-described X-ray CT device 100 includes the functions described above. Of course it is applicable.

In the X-ray see-through device as described above, since the dose is low as described above, the see-through image contains a large amount of random noise caused by photon noise. In order to solve such a problem, if the X-ray irradiation amount is sufficiently increased, the continuous X-ray irradiation is performed, so that the exposure amount to the subject is greatly increased, and the exposure to the inspector is remarkably increased.

By applying the coherent filter according to the present invention, it is possible to solve the above problems and to provide a perspective image in which noise is sufficiently suppressed.

Specifically, the image numbers 1, 2,... , The M is the newest k, i.e. image numbers M, M-1,... , M-K + 1 images are stored in memory, and the k pixels (1 when the present invention is applied to III X-ray CT imaging; dynamic CT imaging (removes noise from multiple images)) The same processing as that of the dynamic coherent filter process or the real-time coherent filter process can be applied.

In addition, although the medical image diagnosis apparatus is far away from the medical imaging apparatus, the above-described method can also be applied to the apparatus used for the so-called nondestructive inspection.

In addition, the present invention can be similarly applied to a tomographic image obtained by an ultrasonic diagnostic apparatus. Also in this case, there is no change in continuously generating images while photographing the subject continuously. Therefore, the coherent filter can be attached and noise can be suppressed by the same configuration as that of the X-ray see-through device.

(Ⅶ More generally when applying the present invention; color image, etc.)

The present invention can also be applied to "device general" or "image general" which deals with images other than the above-described medical image diagnosis apparatus. Hereinafter, an application example of the present invention to such a general apparatus or image will be described. In addition, the description regarding the case where the general "color image" which is an especially important field is made here is mainly performed.

In the color image, as described above, a three-dimensional vector including three brightnesses of light as elements can be configured for each pixel x constituting the image.

In other words,

[Equation 3.1]

* (V _R , V _G , V _B are the three primary colors of light, R (red), G (green), B (blue) light)

to be. As shown in Fig. 2, it can be seen that "corresponding to the pixel value v (x) composed of plural kinds of images".

For example, when photographing the color image with a digital camera composed of a CCD image pickup device, a CMOS image pickup device, or the like, the dispersion of noise included in the pixel value of each pixel is determined by the amount of light incident on the pixel x. do. Specifically, scalar values (V _R (x), V _G (x), V _B (x)) and the variance of noise included in these (σ ² _R (x), σ ² _G (x), σ ² ) _B (x)) are each in almost proportional relationship. This property can be used to obtain an estimate of noise variance. Accordingly, the pixel value V ′ (x) = (V is given by giving the specific form of the same weight function w as in Equation 6, thereby suppressing the noise by configuring the coherent filter in Equation 4. ′ _R (x), V ′ _G (x), V ′ _B (x))).

The term "color" is taken broadly, and in addition to the above-mentioned red, green, and blue, for example, infrared rays, ultraviolet rays, and the like may be included, and may be configured as vectors higher than Equation 3.1, or as vectors not bound to RGB. It may be. For example, it is well-known that the method of expressing a color image using the pixel value of a vector value is not necessarily limited to three primary colors, You may apply this invention to the case where the vector display which does not use such three primary colors is used. . In the case of the four primary colors including ultraviolet (V _u ) and the five primary colors that separate the purple (V _p ) (that is, the time of the butterfly), for example, when the five primary colors, the pixel (x) is V (x) = (V _R (x), V _G (x), V _B (x), V _U (x), V _P (x) are treated as having the five-dimensional vector value, and the same as the above configuration Coherent filter can be attached.

For example, when measuring the light reflection spectrograph of a subject, the wavelength of the illumination light which illuminates a subject is switched step by step or continuously, and it photographs by the monochromatic camera repeated for every wavelength of illumination, or In a photographing apparatus for photographing a multi-wavelength by repetitively photographing by switching the color filter in stages, the pixel value v (x) is selected based on a plurality of these images (short-wavelength or monochromatic), respectively. Can be configured. That is, a coherent filter can be attached by applying the same method as in (V3: (reducing noise of plural kinds of images)) when the present invention is applied to V X-ray CT imaging, resulting in random noise. A plurality of types of images obtained by reducing the number of images are obtained.

In addition, most of the image capturing apparatuses usually perform one shooting in order to obtain one image. However, using the present invention, rather, the same imaging as described in the above-mentioned "X-ray CT imaging" (when applying the present invention to IV X-ray CT imaging; 2 (getting one image with reduced noise))) It is better to repeat the shooting for a short period of time and obtain a plurality of images (k). This is because these k images can be configured (although these k images contain a lot of random noise because the amount of light is reduced by short-time photography). In addition, when these k images are each a color image, for example, you may treat this as "plural types (three types) of images obtained by three primary colors of light are multiple sets (K sets)"). That is, for each pixel x constituting each image, a three-dimensional vector including brightness of three primary colors of light as an element can be configured. Therefore, in all k images, the pixel value of the 3K-dimensional vector (v (x)). ) Can be configured. A coherent filter may be attached to the pixel value of this 3K-dimensional vector.

In addition, the dynamic coherent described in the image capturing apparatus such as a video camera, which is described in (1 when applying the present invention to III X-ray CT imaging; dynamic CT imaging (to obtain multiple images with reduced noise)). Filter processing or real-time coherent filter processing can be applied. As described above, a coherent filter may be attached by treating it as "the plural kinds (three kinds) of images obtained by the three primary colors of light are plural sets (K sets)").

In addition to the above-described digital cameras, the present invention can be applied to images captured by a video camera, a dark camera, a high speed photographing camera, or the like (hereinafter, generally referred to as an "imaging device") in the same manner as described above. In particular, the dark camera and the high speed camera are cameras used in a situation where the amount of light per frame is small, and although there is a lot of random noise included in the image, it is desirable to impair temporal resolution as well as spatial resolution, especially in the high speed camera. Not. Thus, it is one of the most preferred apparatus for applying the present invention.

Therefore, the image pickup device can obtain an image with more noise suppressed by such an imaging method.

On the other hand, the coherent device according to the present invention can be applied to a device capable of reproducing an image based on a signal recorded or transmitted on a certain recording medium, for example, a video recorder or a DVD player. By applying the filter, it is possible to obtain a reproduced image in which random noise is suppressed.

In addition, in the above apparatus, even when generating and displaying a still image (so-called "pose image" or an image for printing), a still image which suppresses random noise is obtained by applying a coherent filter, which is suitable for printing and the like. Do. To this end, the method described in (1 case of applying the present invention to III X-ray CT imaging; dynamic CT imaging (removing noise from a plurality of images)), (3 case of applying the present invention to V X-ray CT imaging); The method described in (Reducing noise of plural kinds of images), the method described later (the pixel value v (x) is composed of one image) or a combination thereof can be used.

(Ⅷ One pixel constitutes the pixel value v (x))

The application examples to the above correspond to what constitutes pixel value v (x) in "plural image" and "plural kind of image" in the classification shown in FIG. In the present invention, in addition to the configuration method of the two pixel values v (x), as shown in Fig. 2, it is also conceivable to configure the pixel values v (x) with one image. It may be carried out as described above.

In a single two-dimensional image, for example, a single CT image acquired by the X-ray CT apparatus 100, an MRI apparatus, or the like or a single image by monochrome imaging by the imaging apparatus, the value of the pixel x is only one. It is composed of scalar values. In other words, the pixel value v (x), which is a vector value, becomes a one-dimensional vector.

In such a case, one method for applying the coherent filter according to the present invention regards the pixel value v (x) as the "one-dimensional vector" itself. That is, in Equation 3, K = 1,

[Equation 3.2]

Shall be. Thereafter, the configuration of the coherent filter or the suppression of the image noise can be performed in the same manner as described above. The present invention includes the case where the "pixel value" (v (x)) as the "vector value" is a one-dimensional vector, that is, a scalar value.

In addition, the following formula (14) can be used for the weight w3 in this case.

[Equation 14]

In Equation 14, D has the same properties as the parameter C of Equation 6. In other words, if D is small, the weight w3 (p (x, y)) is smaller and D is larger.

An image based on such a weight w3 (p (x, y)) and the pixel value v '(x) obtained by the above equation (4) is as shown in Figs. 12 to 16, for example. . 12 to 15 are contrasted images showing 3D extraction examples of blood vessels, and FIG. 16 is an image of brain proteins extracted as another extraction example. 12 shows a CT image (original image) to be processed by the coherent filter. FIG. 13 and FIG. 15 show the results of the coherent filter being cut out using the rectangular region of 3 × 3 pixels as the set N (x) used in Equation 6, and FIG. As a result of setting D-9 in FIG. 15, the result of performing threshold processing on the image in which the coherent filter of FIG. 14 is applied is shown. In addition, in FIG. 14, the image which processed the threshold value with respect to the original image of FIG. 12 is shown.

Comparing the images of FIG. 12 and FIG. 13, it can be seen that the noise of the image of FIG. 13 is adequately suppressed compared to the image of FIG. 12 by applying a coherent filter, and sufficient quality is ensured as a diagnosis.

When the images of Fig. 14 and Fig. 15 are compared, the image of Fig. 14, which has only been subjected to threshold processing on the original image, is difficult to extract blood vessels by the threshold. It can be seen that the image of the blood vessel can be extracted by the threshold value, and no narrow blood vessel is lost.

16 shows that the protein region can be extracted.

By the way, although the effect shown in FIG. 8 thru | or 16 is certainly obtained by such a method, the one-dimensional vector value (v (x) = v (x)) is used to measure the goodness of fit of the pixel (x) and the pixel (y). ), The goodness-of-fit discrimination performance is relatively low, in other words, only a statistically uncertain risk ratio (p (x, y)) is obtained. For this reason, the effect of suppressing random noise which can be performed without compromising resolution is not necessarily much higher. Therefore, in such a case, it is more preferable to adopt another method as follows. That is, a method of constructing a multidimensional vector using a set Z (x) composed of several pixels in the vicinity of the pixel x and using this as the pixel value v (x) of the pixel x is effective. . For example, as an example, as shown in FIG. 17, in addition to the scalar value of the pixel (x = (x, y)) itself, that is, v (x) = v (x, y), the pixel (x) is up, down, left, and right. Using a set (Z (x)) of four pixels (x + 1, y), (x-1, y), (x, y + 1), and (x, y-1) adjacent to Hereinafter, the pixels serving as elements of the set Z (x) around these pixels x are referred to as "z ₁ (x), z ₂ (x), z ₃ (x), z ₄ (x)". Consider a five-dimensional vector to which scalar values of these four pixels are added. In other words,

[Equation 3.3]

to be. Such a vector can be configured similarly to all the pixels constituting the image G shown in FIG. 17 (in addition, FIG. 17 shows only a part of the entire image in an enlarged manner).

In this way, the coherent filter according to the present invention is constructed by using the pixel value v (x) of the multidimensional vector composed of one image, so that the accuracy of the fit is high compared with the case of using the pixel value of the one-dimensional vector. It can be determined. As a result, random noise can be more strongly suppressed without compromising the resolution of the image.

In addition, the null hypothesis H used to quantify the goodness of fit of the pixel in the case where the pixel value is constituted in this way has the same pixel value in the case where the respective noises are removed. Is a null hypothesis (H '), a total of five pixels of the pixel (x) and its surrounding pixels (z ₁ (x) to z ₄ (x)), and the pixel y and its surrounding pixels (z). 5 pixels in total of ₁ (y) to z ₄ (y) each have the same pixel value as that of the corresponding pixel by removing noise ”.

More specifically, as shown in Fig. 17, an arbitrary pixel existing in an arbitrary pixel x and a set Z (x) (this is an area around the pixel x, for example, as described above). When all of (y) exist inside the area E1 of the object E1 shown on the image G shown in the figure, for example, the risk ratio p calculated by Equation 9 or the like, for example. The value of (x, y)) is small, and the value of the weight (w (p (x, y))) calculated by, for example, Equation 6 is large. In this case, the pixels z ₂ (x), z ₄ (x) and the pixels z ₂ (y), z ₄ (y) are in the region of the object E1 and the pixels z ₁ (x ), z ₃ (x)) and the pixels z ₁ (y) and z ₃ (y) are in the image of the object E2, and thus the vector values V (x) and V (y) represented by Equation 3.3. ) Is similar.

In the above embodiment, in particular, when the pixel x in the image E1 of the object is noticed, the pixel y having a small risk p (x, y) when rejecting the null hypothesis H 'is generally used. It is assumed that not many of them can be seen (for example, in FIG. 17, the null hypothesis H ′ is applied even to the pixels x and y _p that exist inside the edge E1E of the image of the object E1. The risk ratio p (x, y) in case of rejection becomes relatively large for the case of pixels (x, y) shown in the figure).

More generally speaking, the value of the risk (p (x, y)) calculated by, for example, Equation 9 is high for most y∈N (x) with respect to the pixel x present on an object. For example, the weight w (p (x, y)) calculated by Equation 6 becomes small. In this case, the smoothing process is weak even when a weighted average is obtained. For example, the noise suppression effect of the coherent filter calculated by Equation 4 is insufficient.

Therefore, another method in which the above method is further improved will be described below. That is, in this method, the pixel value v (x) is

[Equation 3.4]

It consists of. However, the expression of v ₁ 3.4 ₄ ~v is v (x + 1, y) of the equation 3. 3, v (x-1 , y), v (x, y + 1) and v (x, y-1) is reordered according to a certain rule (for example, in order of increasing value).

In this way, the null hypothesis H 'corresponds to a total of five pixels of the null hypothesis H''"pixel (x) and its surrounding pixels z ₁ (x) to z ₄ (x)). A total of five pixels of the pixel y and the surrounding pixels z ₁ (y) to z ₄ (y) are respectively shown in [pixel z ₁ (x) to z ₄ (x) and z ₁ (y) to allowing the input to change the order within z ₄ (y)], but having the same value except for noise ”(indicated by [].).

In this case, the null hypothesis H '' is rejected even in the relationship between the pixel x and the pixel y existing inside the edge E2E of the image E2 of the object shown in FIG. 18, for example. The risk (p (x, y)) becomes small (because pixels z ₂ (x), z ₃ (x), z ₄ (x) and z ₁ (y), z ₂ (y), z). ₄ (y)) matches v ₁ to v ₃ in the above equation 3.4, and the pixel z ₁ (x) and the pixel z ₃ (y) identically match v ₄ as a result. The noise suppression effect is larger than that by the configuration of the pixel value v (x) according to Equation 3.3, and the spatial resolution is not impaired.

As described above, in the present invention, the pixel value v (x), which is a vector value, can be constituted by one image. Therefore, the coherent filter can be used to suppress noise without damaging the spatial resolution. Since such a system needs only one image, it is generally applicable to any image. That is, this method can be applied to all the images as well as the above-described various application examples (X-ray CT apparatus, medical image diagnosis apparatus or color image, etc.). In addition, even when an image is a moving image (as described many times above, since it can be viewed as a large number of still images), the method of viewing a single image of the still image ("Pixel value (v (x) Of course, you can apply)). In addition, when using the present method to an image of a three-dimensional or multi-dimensional (Z (x)), for example, such that can be comprised of six pixels and χ ₁ itself in contact with the χ _1, dance be very easy way to extend You don't need to explain it.

In addition, in the present invention, the pixel values v (x) constituted by the independent methods of &quot; plural images &quot;, &quot; plural kinds of images &quot; or &quot; single images &quot; Of course, a coherent filter may be configured, but each of these methods may be appropriately combined to form a pixel value v (x) represented as a vector value of a higher dimension, and based on the coherent filter. It is also possible to construct a filter.

As already described in Fig. 2, for example, if there are two color images of the same subject, the scalar values v _R (x), v _G (x) and v _B (x) is the pixel value v (x) of the pixel x in each of two images,

[Equation 3.5]

You can get a six-dimensional vector called. When the pixel value v (x) represented by a higher dimension vector is used as described above, a more certain value is obtained as the risk (p (x, y)) calculated by, for example, Equations 9 and 6, and the like. Can be calculated. As a result, it is possible to more strongly suppress random noise without compromising the resolution of the image.

(Ⅸ Most common form of coherent filter of this embodiment)

In the following description, a more general form of the coherent filter introduced as shown in Equation 4 is described.

In the general form of the coherent filter (Equation 4) or various application examples, the null hypothesis H described first "the pixels x and the pixels y have the same value when the respective noises are removed". To &quot; v (x) = v (y). However, a difference due to noise of both pixels is excluded '' can be used as the general expression shown below.

That is, a model function f (a, v (y)) including an unknown parameter of M dimension (a = (a ₁ , a ₂ ,..., A _M ) (where M <K)) is introduced. In the two vectors (x, y) indicating the position of the pixel, the null hypothesis (H ₀ ) &quot; v (x) = f (a, v (y)) + ξ (where ξ is ), According to the probability distribution. At this time, the estimated values (a) of the parameter (a) are determined by using a suitable fitting method (e.g., least square method, described later) in v (x) and v (y). You can change the null hypothesis to the equivalent proposition H _0, "v (x) = f (a ~, v (y)) + ξ, where ξ depends on the (probability) probability distribution." Here, f (a to v (y) is the degrees of freedom allowed by the model describing the function f with respect to the pixel value v (y) of the pixel y, that is, the parameter (a). ) Is optimally adjusted and converted to have the highest suitability with the pixel value v (x) of the pixel x. And such optimal parameter is a ~.

As a result, the risk (p (x, y)) can be specifically calculated using a known probability distribution (where the ξ is assumed if the null hypothesis is correct), and thus the weights w (p (x , y))) can be calculated, so the weighted average (v ′ _k (x))

[Equation 15]

Can be calculated by This is the format of the most common coherent filter of this embodiment. Also, if the equation (4) and the equation (15) are compared, the former is actually the latter.

[Equation 16]

I did. It can be seen that it must be a more specific form.

(Ⅸ-1 Application Example Based on Equation 15; Configuration of Time Concentration Curve Based on Dynamic Phase)

Another application example of the present invention using the format shown in equation (15) of the coherent filter will be described. In this application example, when the quantitative measurement of a time-density curve for a dynamic CT image obtained by the medical image diagnosis apparatus is performed, the present invention is applied to sufficiently suppress random noise. The Example which aims at obtaining the said time concentration curve is demonstrated.

First of all, the term "dynamic image" referred to herein means a series of images (image numbers 1, 2, ..., K) obtained by repeatedly photographing the same subject, and respective shooting times (t ₁ , t ₂ , ..., t). _K ), for example, a dynamic CT image obtained by the medical imaging apparatus, that is, the X-ray CT apparatus 100, an X-ray perspective apparatus, a scintillation camera, an MRI apparatus, a SPECT apparatus or a PET apparatus, The image is taken using an ultrasonic diagnostic apparatus. The dynamic image may be configured by using a series of images obtained as a result of repeated photographing by the above-mentioned general photographing apparatus, that is, a digital camera, a video camera, a dark camera, a high speed photographing camera, or the like.

In addition, the "time concentration curve" used here is a curve which shows the time-dependent change of the density | concentration value of the phase of the specific site | part in the said dynamic phase. In particular, the medical imaging apparatus measures time-dependent changes such as contrast agent concentrations in specific tissues of the human body as a time concentration curve for the purpose of detailing blood flow dynamics and metabolic functions of human tissues. In the celestial observation, etc., the time concentration curve is used for the purpose of analyzing the luminosity change of the specific celestial body. More formally, that is, the time concentration curve means that when the concentration value of any part of time t _k is d _k , the pair of columns {<t _k , d _k > (k = 1, 2,... , K)}. In addition, in most applications of the time concentration curve, an absolute value of d _k is not necessarily required, but only an increment d _k -d ₁ based on the first image 1 is sufficient. In addition, many of these same purposes only (d _k -d ₁₎ a data A (d _k -d ₁₎ proportional to it is sufficient is obtained only (here A is an unknown coefficient proportional). In this case, therefore, the pair of columns {<t _k , A (d _k -d ₁ )> (k = 1, 2, ..., K)} is a time concentration curve.

In order to obtain such a time concentration curve, in principle, the pixel (x) included in a portion where the time concentration curve of each image k constituting the dynamic image (k = 1, 2, ..., K) is to be measured. Column using the scalar value (v _k (x)) of the phase {<t _k , v _k (x)>} or {<t _k , A (v _k (x) -v ₁ (x))>} It is good to configure.

However, in practical use, since the random noise is included in the dynamic image photographed by the medical image diagnosis apparatus or the like, there is a problem that the time concentration curve originally intended to be measured cannot be accurately obtained.

Moreover, in practical use, what is called a "personal volume effect" arises in these dynamic phases. The physical borium effect, i.e., the image of a small object in the subject is represented by a few pixels on the image, but the pixel value of these few pixels is affected because the small number of pixels affects the image of adjacent objects in the subject. This is a phenomenon in which only relatively small fluctuations appear (although proportional to fluctuations in concentration values to be originally measured). In other words, the pixel values of these few pixels contain only a few signals. Therefore, in the case where the physical borium effect occurs, the pair of columns {<t _k , v _k (x)> (k = 1, 2, ..., K)} has a very low signal level no matter what pixel x is taken. Due to a change in the concentration value of the tissue, which is not intended to be measured originally, and random noise exist, there is a problem that the time concentration curve {<t _k , d _k >} to be originally measured cannot be accurately obtained.

Therefore, conventionally, in order to suppress random noise, first, a pair of columns {<t _k , v _k (x)> (k = 1, 2, ..., K) relating to the pixel x included in the portion to be measured. } And then find the time average (v _k # (x))

[Equation 17]

And time average value of the corresponding time,

Equation 18

(Where m <n is an appropriate integer and G _j is an appropriate weighting factor), constructing a time concentration curve {<t _k #, v _k # (x> (k = 1, 2, ..., K)} The time-based method using this as a time concentration curve is used.

Alternatively, conventionally, a set R of pixels (ie, a region of interest (ROI) on an image) corresponding to a portion to be measured is set in order to suppress random noise. The time average value v _k (R) of the scalar value v _k (x) of all the pixels x ∈ R included in the set is obtained by the following equation (19), and the columns of the pair {<t _k , v _k (R) &gt; (k = 1, 2, ..., K)}, and a method using a spatial average that uses this as a time concentration curve is used.

[Equation 19]

Moreover, the method which suppresses more random noise by using together time average and space average is also used.

However, according to these conventional methods, there is a problem that the time resolution is lost when the time averaging is performed, and that the time-dependent change in the concentration of the parts other than the original site to be measured is mixed in the measured value when the time averaging is performed.

In order to solve such a problem and to obtain a more accurate time concentration curve, the coherent filter according to the present invention can be applied based on the above (the most common form of the coherent filter of the present embodiment).

First, the null hypothesis to be used in the coherent filter of the present embodiment will be described. Assuming that the true time-concentration curve of the part to be measured is {<t _k , d _k > (k = 1, 2,…, K)}, the first transform {<t _k , A (d _k -d ₁ )> (k = 1, 2, ..., K)} (where A is an unknown coefficient), the set of pixels R corresponding to the area to be measured is set. . Condition Q for any pixel x∈R that is an element of this set R: &quot; If this pixel x reflects the true time concentration curve well, the concentration change over time in another portion Is hardly affected by the effect of the perical volume effect on the pixel value (as a vector value), v (x) = (v ₁ (x), v ₂ (x), ..., v _K (x)) By considering the effects of random noise,

[Equation 20]

It can be assumed that this holds true. Here, p (x) and q (x) are unknown coefficients that differ for each pixel x but do not change with the image number k (i.e., the shooting time t _k ), and model the physical volume effect. It is. Γ _k (x) is a model of random noise. The value is different for each pixel x and for each image number k, but the expected value is 0. k). Also, for the sake of explanation, an example in which the statistic distribution is a normal distribution with an average of 0 and a variance (σ ² ) will be used, but the statistic distribution is roughly determined by any known distribution. What is necessary is just to be approximated, and in that case, the method of modifying the following embodiment is clear.

According to the above assumptions, for any two pixels (x, y) that are elements of the set (R), if "the pixels (x, y) both satisfy the condition (Q) above". If it is true, it can be proved that the relationship of Equation 21 below holds.

[Equation 21]

Here, a ₁ and a ₂ are unknown coefficients that differ for each set (x, y) of the pixel but do not change depending on the image number k (that is, the shooting time t _k ). Note that ξ _k is random noise, and the value is different for each set (x, y) of pixels and for each image number k, but the expected value is zero.

Equation 21 is derived as follows. That is, the equation obtained by substituting y to x in Equation 19

[Equation 22]

Transforming

[Equation 22 ']

Is obtained by substituting

[Equation 23]

Get therefore

[Equation 24]

(21) is derived. Here, a ₁ and a ₂ in Equation (24) are parameters representing the partial borium effect, and ξ _k in (24) represents random noise.

In the above, the proposition that "the pixels (x, y) both satisfy the condition (Q)" is the null hypothesis (H ₀ ′) "v _k (x) = a ₁ v _k (y) + a ₂ + ξ _k (k = 1, ..., K). "

Subsequently, the null hypothesis (H ₀ ′) is substantially equivalent and actually tested, ie, v _k (x) = a ₁ v _k (y) + a ₂ + ξ _k (k = 1, ..., K). Describes how to convert a proposition to a format that can be done. This null hypothesis is explained in mathematically rigorous terms: the null hypothesis (H ₀ ′) “any integer (a ₁ , a ₂ ) exists, ξ _k = v _k (x) -a ₁ v _k (y) -a ₂ (k = 1, ..., K) follows a normal distribution of mean 0 and variance (σ ² h (a ₁ )). It becomes. The coefficient h (a ₁ ) is

[Equation 25]

to be. Equation (25) is directly derived because a general nature are distributed about a ₁ and definition of equation (24) and a random variable of ξ _k with.) In addition, the value of the variance (σ ²⁾ is the one (Ⅲ X-ray CT imaging In the case of applying the present invention to the present invention, the method can be estimated simply and practically sufficiently accurately by the method described in 1; dynamic CT imaging (to obtain a plurality of images with reduced noise).

Thus, if the integers a ₁ and a ₂ can be determined, it is possible to test the null hypothesis H ₀ ′. In reality, it is sufficient that the optimum estimated values (a ₁ to a ₂ to) of these integers are obtained.

Such a constant (a _1, a ₂₎ the calculation of the best estimate may be used as it is a known applicable law (fitting). Therefore, below, the outline | summary in the case of using the linear least square method as a typical specific example of such an application method is demonstrated. Applying the linear least squares method to this embodiment only assumes that the sum of squares of ξ _k of the null hypothesis is S (a), i.e.

[Equation 26]

Define. The value of S (a) depends on the integer vector a = (a ₁ , a ₂ ), i.e. the value of said integer a ₁ , a ₂ . When this S (a) calculates the integer vector (a) which takes the minimum value, the optimal estimated values (a ₁ to a ₂ to) in the sense of inconvenient estimation for the constants (a ₁ , a ₂ ) are obtained. Obtained. As a specific calculation method of the linear least square method, various known methods can be used, and all of these known calculation methods are very simple, and the required calculation time is very small.

As a result of calculating the optimum estimated values a ₁ to a ₂ to the integers a ₁ and a _{2 in} this manner, the residuals defined in the following equation (27):

[Equation 27]

Can be specifically calculated. Thus, the residual (r _k ~) the null hypothesis (H ₀ ') a substantially equivalent null hypothesis (H ₀ ") using the" _{r k ~ (x, y)} (k = 1, ..., K) Depends on a normal distribution of mean 0 and variance (1+ (a ₁ ˜ ² ) σ ² ). This is actually a specific proposition that makes the calculation of the test feasible.

Also, a representation by vector

[Equation 28]

(However, the vectors (a) and ξ depend on the set of pixels (x, y)).

[Equation 29]

The null hypothesis in other words, the (H ₀ ') the null hypothesis (H ₀ ") using a vector function (f), defined as the" v (x) = f (a ~, v (y)) + ξ, ( stage , ξ is an average of 0 and the variance (following the normal distribution of 1+ (a ₁ ˜ ² ) σ ² ) '', which is the null described in (the most common form of coherent filter in this embodiment). It is in exactly the same form as the hypothesis (H ₀ ). That is, it is clear that this embodiment is one modification of the coherent filter which concerns on this invention. Here, f (a to v (y)) means that the parameter (a) which shows the effect of the partial volume on the pixel value v (y) of the pixel y is optimally adjusted. This means that the pixel value v (x) of x is converted to have the highest goodness of fit.

Subsequently, a description will be given of a method for obtaining a time concentration curve by a coherent filter using the null hypothesis H ₀ ″ in the present embodiment. With respect to the set R of pixels approximately corresponding to the portion to be measured, all pixels y_R that are elements of the set R with respect to any one pixel x_R included in this set R The following calculation is performed. That is, the residual (r _k ˜ (x, y)) (k = 1,…, K) is actually _calculated using the above method, and then the null hypothesis (H ₀ ”)“ r _k ˜ (x, y) is calculated. (k = 1,…, K) is based on the normal distribution of mean 0 and variance (1+ (a ₁ ～) ² ) σ ² '' risk (p (x, y)) to weight (w (p (x, y))) is specifically calculated by the following equation. Then, the weighted average (v time density curve {<t _k, v _'k (x) -v' ₁ (x)> of the calculation, and the pixel (x) by _"k (x)) to the equation (15) ' (k = 1, 2, ..., K)}.

[Equation 15 ’]

The time concentration curve thus obtained is {<t _k , A (d _k -d ₁ )>}, where A is the first-order transformation of the actual time concentration curve {<t _k , d _k >} of pixel x, where A is unknown. Coefficient) is a measured value approximating, and random noise is suppressed by the effect of the weighted average by (15). As for the pixel value vector of the other pixel y used for the calculation by the equation (15), the correction of the influence of the physical volume effect is used as is apparent from the equation. In addition, the present embodiment has the property of `` not using all of the time average and calculating the spatial average using a weight based on the goodness of fit of the pixel x '', which is a common feature of the coherent filter. Therefore, according to the present embodiment, it is possible to obtain a time concentration curve in which the influence of the physical borium effect is suppressed and random noise is suppressed without impairing the time resolution. In this way, the method for obtaining the time concentration curve is particularly referred to as the "coherent relaxation method."

Subsequently, an example of clinical use of the dynamic phase concentration curve obtained by dynamic CT imaging of medical X-ray CT will be described in detail. In this application example, the blood flow dynamics of the tissue are diagnosed by measuring the concentration change of the arterial phase present in the human tissue as a time concentration curve by performing dynamic CT imaging while rapidly injecting the contrast agent into the blood vessel.

In most cases of this application, the arteries in human tissue are generally very thin, so the image of the arteries appearing on tomographic images by CT causes the effect of the boron. It goes without saying that random noise is contained in the image. For this reason, in the conventional method, it is difficult to obtain a sufficiently accurate time concentration curve for an artery, and if a strong measurement is performed, the first conversion of the actual time concentration curve <t _k , D _k > for the artery is <t _k , A (D _k). a -D _1)> (here D _k is the time (t _k of the pixels of the group corresponding to the image of the artery) indicates the (scalar value) of pixel values. also, k (= 1, 2, ..., k) Only a somewhat approximate measurement (<t _k , (v _k (x) -v ₁ (x))>) was obtained, which included random noise, and because of the effect of the physical volume effect, (A) remains unknown.

Therefore, when the above method according to the present invention is applied, the measured value <t _k , (v ′ _k (x) −v ′ ₁ (x)) &gt; sufficiently approximating &lt; t _k , A (D _k -D ₁ ) &gt; k = 1, 2, ..., K) can be obtained. On the other hand, some of the veins that can be observed on the same tomography image are considerably thicker. Therefore, as for those veins, an approximate value <t _k , (J _k -J ₁ )> (k _k -J ₁ )> (k k) having a sufficiently good time concentration curve by the conventional method is k. = 1, 2, ..., K) can be obtained. J _k here represents a pixel value of the time t _k of a group of pixels corresponding to the vein image.

However, in the time-concentration curve for blood circulation, the proposition S: `` If the concentration of contrast medium in the blood at time (t ₁ ) is 0, the time-concentration curve for certain vessels (d _k , (d _k -d ₁ ) It is known that &quot;&quot; matches the area under curve (AUC) &quot;. Here, the area under the curve means the integral of the time t of the time concentration curve <t _k , (d _k -d ₁ )>.

Therefore, the area under the curve AUC (d) of the time concentration curve <t _k , (d _k -d ₁ )> for a certain blood vessel d can be approximately calculated by, for example, equation (30).

Equation 30

Therefore, the area under the curve AUC (J) relating to the time concentration curve {<t _k , (J _k -J ₁ )>) obtained by the conventional method with respect to the vein can be calculated using the above equation (30). (It is good to substitute J for d.) Also, if the time concentration curve {{t _k , (D _k -D ₁ )>) is known about the vein, the area under the curve (AUC (D)) is expressed by Equation 26. Can be calculated in the same manner, and in accordance with the proposition S,

Equation 31

This holds true. In practice, however, the time concentration curve <t _k , (D _k -D ₁ )> is unknown, so AUC (D) cannot be calculated.

On the other hand, the time concentration curve <t _k , (v ' _k (x) -v' ₁ (x))> obtained by the method according to the present invention approximates <t _k , (D _k -D ₁ )>, The latter contains an unknown coefficient (A). For this reason, the area under the curve (AUC (v ′) is AUC (D) which can be specifically calculated using Equation 30 in {<t _k , (v ′ _k (x) −v ′ ₁ (x)))}. Must not be exactly A times

Equation 32

to be. That is, from equation (31) and equation (32),

[Equation 33]

Relationship is established. Since the right side of Equation 33 can be specifically calculated using Equation 30, it is possible to specifically determine the value of the unknown coefficient A. Therefore, if the time concentration curve <t _k , (v ' _k (x) -v' ₁ (x)) / A>) is constructed using the value of this coefficient (A), then the time concentration curve <t _k , (D _k -D ₁ )> must be approximated. Thus, the method of constructing the time concentration curve which determined the value of the unknown proportional coefficient A using the area under the curve is called "AUC method".

By combining the AUC method in addition to the coherent relaxation method in the clinical use of the dynamic phase concentration curve obtained by dynamic CT imaging and the like, the measurement of thin arteries was difficult or impossible with conventional methods. Regarding the time concentration curve, a measurement value is obtained which excludes the influence of the physical volume effect and random noise, and does not contain an unknown proportional coefficient (A).

Of course, the AUC method can also be applied to a time concentration curve <t _k , (v ′ _k (x) -v ′ ₁ (x))> for arteries measured by a conventional method alone, and (random The effect of noise and the physical and volume effects can be excluded), but a time concentration curve that determines the value of the unknown proportional coefficient (A) can be constructed.

The time concentration curve obtained by combining the AUC method in addition to the coherent regression method is represented by a symbol A in FIG. 19 for an arbitrary pixel x, for example. In addition, this figure also shows that (B) which applied the said AUC method independently to the time concentration curve comprised by the conventional method. While the effect of random noise is clearly seen in (B), it is understood that the noise is sufficiently suppressed in (A) and that the time resolution is not impaired at all.

In addition, although the scalar value _dk was used in the above description as a concentration value of a time concentration curve, this embodiment is not limited to this case, For example, each image which comprises a dynamic image is a color image, In general, it is a set of various kinds of images, and of course, it can be easily extended and applied even when the concentration value of the time concentration curve is expressed as a vector value.

(Ⅹ Supplemental matter of this embodiment)

Hereinafter, the supplementary matter of embodiment described above is demonstrated.

(Ⅹ-1 General Supplements)

First, in the above embodiment, the "risk rate" (p (x, y)) in the case of rejecting the hypothesis H as a means of quantifying the "fit" in the present invention is assumed. It is not limited. As described first, "fit" may be a numerical "indicator" indicating whether the pixels (x, y) are similar in some sense.

In addition, in the above description, the χ square test method is used to calculate the risk in the case of rejecting the various null hypotheses (see Equations 6 and 9), but the present invention is not limited thereto. In other words, this invention is not limited only to the form using the "statistical test method" or the "(square) test method which is a kind of specific form.

In the above embodiment, as a specific form of the weight w (p (x, y)), two kinds of weight functions (w1 (Equation 10 to which Equations 6 to 9 are substituted)) and Although w3 (Equation 14) is mentioned as an example, this invention is not limited to this. As already explained, the property that the weighting function (w) as a function of the risk (p (x, y) ∈ [0, 1]) must satisfy is that w (t) is defined as the domain (t∈ [0, 1]). It is just a monotonic increment function. Most generally speaking, the weight function w may be a "non-negative monotonic increasing function related to fitness".

(X-2 more general form of goodness of fit)

The "fit" referred to in the present invention is not limited to the method of quantifying by the risk ratio p (x, y) when rejecting the null hypothesis described above. Therefore, the "weight" is not limited to the method of calculating the weight function w by acting on the risk ratio p (x, y). Hereinafter, a description will be given according to a specific example with respect to a more general form of such a goodness of fit.

First, in general, the present invention digitizes in a suitable manner according to the purpose that certain propositions appropriately set in accordance with the purpose of the image processing are established in the image to be processed, and that numerical values are applied to the image processing. Take the configuration to apply. For example, in order to quantify the goodness of fit of the pair of pixels (x, y), a gentler scale may be constructed in the present invention. As such a specific example, a configuration such as quantifying the goodness of fit using a membership function of fuzzy logic can be considered.

(Ⅹ-2-1 Example of applying the present invention to pattern recognition)

Here, an example in the case of applying the present invention to the so-called "pattern recognition" which identifies and extracts the characteristic pattern on an image as a specific example of a more general form regarding fitness, etc. is demonstrated.

In this case, a function (m (x)) indicating the certainty of the proposition "pixel (x) is a pixel constituting the pattern" based on the characteristics of the pattern to be identified in advance is defined. However, m (x) ∈ [0, 1] is assumed. More specifically, m (x) is each pixel y included in the vector value v (x) which is the pixel value of the pixel x and the set N (x) of pixels in the vicinity of the pixel x. Is a function of calculating the numerical value representing the certainty of the proposition by the pixel value v (y) of the vector and the vector x indicating the position of the pixel itself. Therefore, m (x) needs to be appropriately designed according to the difference between the pattern to be extracted and the pattern not to be extracted. Also, the weight function

[Equation 34]

Defined as And if this m (x) comprises the set X of pixels more than arbitrary "threshold value", the area | region of the pixel x corresponding to a pattern will be obtained.

When the image P to be processed is given, w (T, m (x)) is obtained by substituting the specific threshold value T and the m (x) into the weighting function for all the pixels x of the image. (This sets the set X (that is, the set of pixels x with w (T, m (x)) = 1) with w (T, m (x)) as a characteristic function.) This X must be an area occupied by the pattern on the image P). In addition, a new image P 'is generated as follows. That is, the pixel x of P 'has v (x) as the pixel value when w (T, m (x)) = 1, and 0 (zero vector) as the pixel value otherwise. The image P 'thus produced is obtained by erasing pixel values of pixels other than the pattern to be identified in the image P by zero.

Of course, the calculation of m (x), the calculation of w (T, m (x)), and the processing of the configuration of P 'may be performed respectively, or the calculation may be performed by software incorporating these processes. Do.

In this example, the function m (x) combines the characteristics of the pattern with the vector value v (x), which is the pixel value of the pixel x, and the value of x itself, to &quot; a goodness of fit &quot; Digitize Then, the weights w (T, m (x)) are calculated by the weight function defined in equation (34), and a new image P ', which is the result of the processing, is formed using the weights and the image P. .

As a specific example of this, a configuration of an image processing for extracting fine lines reflected in an aerial photograph, a micrograph of a metal surface, or the like will be described.

First, the method of constructing the index numerically indicating the certainty of "the pixel (x) which constitutes a thin line" is demonstrated.

In advance, a set N (x) of pixels in the vicinity of the pixel x is defined. Typically, N (x) may be defined as a rectangular region centered on the pixel x. Then, the multidimensional vector value v (x) which defines the pixel value (scalar value or vector value) of each pixel of N (x) is defined. Let K be the dimension of this vector. Typically, K is preferably about 25 to 169.

Subsequently, the normalization function (nrm), a K-dimensional vector function that is used as an aid to subsequent calculations, is defined as follows. That is, for any K-dimensional vector (v) with v ≠ 0, the normalization function (nrm) is

[Equation 35]

Provided that k = 1, 2,... , K. With this normalization function, for any K-dimensional vector (v) (unless v is a zero vector),

Is guaranteed.

Subsequently, one pixel x is arbitrarily determined, and a typical example of an image having only one thin straight line passing through the pixel x as an image having a background of 0 is collected by only a suitable number J. Image numbers 1, 2,... , J is added. By calculating the value of v (x) of the image of the image number j, the vector r (j) is

It comprises concretely by. This calculation is j = 1, 2,... For J. (The J vectors r (j) constituted as described above will be referred to as &quot; pattern vectors &quot; hereinafter. The pattern vectors correspond to &quot; pixel values composed separately &quot; as used in the present invention.)

In addition, the function m (x) is defined by the following equation.

Here, "*" shows the inner product of K-dimensional vectors, and "max" is a function which takes out the maximum value of a set. The function m (x) thus constructed takes a large value when the pixel x matches or is similar to any one of the pattern vectors r (j) (j = 1, 2,..., J), otherwise Since it takes a small value, it becomes a function of numerically indicating the certainty of the proposition that "pixel (x) is a pixel constituting a thin line", that is, a function of "conformance."

(Equation 34)

Defined as Here, T is an integer that becomes a threshold, and an appropriate value is set.

When the image P to be processed is given, w (T, m (x)) obtained by substituting the specific threshold value T and the m (x) into the weight function for all the pixels x of the image is given. Calculate (This defines a set X (that is, a set of pixels x with w (T, m (x)) = 1) with (w (T, m (x)) as a characteristic function), This X must be an area occupied by the pattern on the image P.)

In addition, a new image P 'is generated as follows. That is, the pixel x of P 'has v (x) as the pixel value when w (T, m (x)) = 1, and 0 (zero vector) as the pixel value otherwise. The image P 'thus produced is obtained by erasing pixel values of pixels other than the pattern to be identified in the image P by zero.

In addition, an ambiguous determination function of w (T, t), for example

Where C is a positive integer

The pixel x of P 'may be composed of w (T, m (x) V (x) with pixel values, which is an image that is judged not to correspond to the "thin line" part by obscure logic. By lowering the contrast, the effect of constructing the image P 'so as to lift the "thin line" portion is produced.

(Supplemental Notes on the Estimation Method of Ⅹ-3 Variance (σ ² ))

Subsequently, a supplementary explanation about a method for estimating the variance σ ² (to the standard deviation σ) used in the above Equation 6 or 14 will be given.

In the case of the "applied to X-ray CT imaging", because of the background that the expected value of dispersion (E [σ ² ]) obtained by the above equations (6) and (7) is valid for all pixels (x) on all k images. the art all of the pixels (x) uniformly in the art per word was being used in the form of E [σ ^2]. However, if such a premise is not satisfied, i.e. for each pixel x (here, x ₁ , ..., x _F ), the inherent dispersion (even if the distribution type is the same (e.g. Gaussian distribution)) sigma ² (x ₁ ), ..., sigma ² (x _F ) are generally assumed.

In this case, the expected values E [σ ² (x ₁ )],…, E [σ ² (x _F )] are obtained separately for the respective pixels x by the above equations (6) and (7). You can do it. Thereafter, the expected values E [σ ² (x ₁ )],…, E [σ ² (x _F )] obtained separately in this way are used in each case (= pixel (x ₁ ,... x _F )), it is substituted into Equation 3, and v '(x) = (v' ₁ (x), v ' ₂ (x), ..., v' _K (x)) may be obtained, respectively.

* In addition, in the case where the noise contained in the pixel values of all the pixels is assumed to be said to have a common variance (σ ²⁾ It is also possible to employ the following methods to carry out the estimation of the variance (σ ^2). First, the expected value of the variance E [σ ² (x)] obtained by the above equations (7) and (8) with respect to the set of all pixels x (Ω (= {x ₁ , ..., x _F })). ) to obtain an average ((σ ^{2) +),} which is taken as a first approximation of the estimate of σ ² of the equation (6) of the. In other words,

to be. However, in the above equation, |? | Indicates the number of elements of the set?, That is, |? | = F. Subsequently, E (σ ² (x))] sets M as a set of pixels x having, for example, several times or less of the average value (σ ² ) ⁺ . If (σ ² ) ⁺⁺ is averaged for only the pixels (x) that seem to be elements of M,

[Equation 36]

Becomes And this can be used as a more plausible estimate of σ.

This is specifically the case shown next, i.e., as shown in Fig. 20A, there are two still images G1 and G2 constituting a moving image, but as shown in Fig. 20B. The difference between (G1, G2) is effective when the images Z1 and Z2 are slightly displaced from each other (that is, the image Z1 moves like the image Z2).

The variance σ ^{2 +} in Equation 35 corresponds to the variance obtained based on all the pixels x (= Ω) constituting the image G3 shown in FIG. 20B in this figure. However, in this case, as shown in Fig. 20 (c), in the probability distribution diagram of the image G3, an example is caused by the portions ζ ₁ and ζ ₂ where the images G1 and G2 do not overlap each other. For example, since two acids (ζ ₁ ′, ζ ₂ ′) are produced, the variance σ ^{2 +} is overestimated, and thus it is inappropriate to use it for the whole still image. On the other hand, FIG. 20 (b) for a part (ζ _M) shown in the middle acid of Figure _{20 (c) (ζ M '} ) is therefore the idea that the acid (ζ _M') to leave said two acid ( It is preferable to obtain the variance of the established distribution diagram of the state except ζ ₁ ′, g 2 ′). And only such a variance corresponds to the variance (σ ² ) ⁺⁺ of Equation 32 above.

That is, the average of "E [σ ² (x)]" of only the "set M" of pixels x within a multiple of (σ ² ) ⁺ is averaged in the acid (Fig. 20 (c)). It is not a process of taking only ζ _M ′) and taking the average, and as a result, (σ ² ) ⁺⁺ yields the variance of the established distribution of the state excluding the acid (ζ ₁ ′, ζ ₂ ′) of FIG. 20 (c). do.

Therefore, in this case, by substituting this (σ ² ) ⁺⁺ into the above equation (5) or the like, the pixel value after conversion for each pixel (x ₁ , ..., x _F ) (v '(x) = (v' ₁₎ (x), v ' ₂ (x), ..., v' _K (x))) can obtain more plausible values.

In addition, the method of estimating the noise is generally a method of estimating the distribution of noise by placing a subject of constant brightness at a position reflected in an edge of an image and measuring the distribution of a value of a set of pixels corresponding to this portion. Is also valid.

In the present invention, the method of estimating the distribution or variance of noise may be basically any method, and is not limited to some methods described herein. In practice, it is necessary to adopt a method that is appropriately and most appropriately obtained by using a priori information, an imaging process theory, and measured to actual values available in various embodiments. In this case, the estimation method is preferably configured to be as simple as possible in accordance with the practical precision required in the examples, in order to simplify the configuration of the apparatus and to improve the processing speed.

(X-4 Another Embodiment of the Present Invention)

Finally, another embodiment of the present invention not mentioned above will be supplemented.

(Example as X-4-1 Data Compression Preprocessing)

In general, if an image Q 'is added with random noise to an arbitrary image Q, the image Q' has a much larger amount of information than the image Q (except the original image Q is very large). If it contains noise). Therefore, if such an image Q 'is to be recorded on a communication or recording medium, the amount of information for describing noise having no practical meaning wastes the number of bits of the communication or recording. Conversely, if the image Q 'is suppressed from the image Q' and the object is a communication or recording object, the image having the same meaning (and it is clear) can be represented with a small amount of information. .

In general, when an image is the object of communication or recording, the image is often subjected to so-called "compression processing". This aims to enable high speed communication or mass recording (for one recording medium) by making the amount of information to be transmitted or recorded as small as possible. In practice, it is especially the case of a "movie image" that the information amount is desired to be small. These are well known facts in irreversible picture compression techniques.

Based on the above matters, the compression of the image Q 'with the noise suppressed may reduce the number of bits (i.e., the amount of information) required for communication or recording, rather than the compression of the image Q' containing noise. Can be.

Therefore, the preprocessing (specifically, the "preprocessing apparatus") which attaches the coherent filter which concerns on this invention, and suppresses the noise with respect to the image Q 'to communicate or record is comprised by hardware or software. The same) is obtained.) Image Q '' obtained by suppressing noise without compromising temporal resolution and spatial resolution is obtained, and the image Q '' can be compressed at a high compression rate, and communication and recording are possible. The time and memory required for this can be greatly saved.

(Example as Pretreatment of X-4-2 Two-Dimensional or Three-Dimensional Region Extraction Processing)

A process of extracting a specific region from a two-dimensional image such as an aerial photograph or a three-dimensional distributed image by the MRI apparatus or the like is often performed. For example, in the former aerial photograph, a region of only a road is extracted, or a region of only a blood vessel is extracted from a three-dimensional distribution image of the human head by the latter MRI device, and the like is rendered and displayed as a computer graphic. In particular, in the imaging of the latter vessels, it can be rotated or observed in various directions, and is recognized as an important technique in diagnosing cerebrovascular disorders.

In the case of performing such region extraction, the simplest and basically executed process is so-called "threshold processing". In other words, the threshold value processing refers to a process of classifying each pixel x according to whether or not the scalar value of each pixel x constituting the two-dimensional image, the three-dimensional distributed image, or the like is greater than or equal to an arbitrary threshold value. . If a set of pixels consisting of only pixels x that fall into a specific classification is rendered, an image that has undergone the region extraction, that is, a "vessel shape only for blood vessels," for example, can be represented (although in the prior art, "threshold processing" is also known. There are many other ways to use.

By the way, in this area extraction processing, it is preferable that the noise of the image (the above-mentioned two-dimensional image or three-dimensional distributed image, etc.) as a background is small. If there is noise in the image, for example, a pixel in a place other than the original blood vessel has the value of a pixel as if it is a blood vessel, and conversely, a pixel in a place where the original blood vessel is not a blood vessel is obtained. This is because the state of having it occurs. Therefore, in this case, a process of removing such misclassified pixels by complicated post-processing is required. In the development, generally, the "smoothing process" described in the prior art section is generally performed to remove the noise on the underlying image, but in this case, the spatial resolution is impaired and the thin structure of the region to be extracted ( For example, there is a problem of losing the blood vessels).

Therefore, if the coherent filter according to the present invention is attached to the underlying image, that is, the secondary image or the three-dimensional distributed image, and preprocessing is performed to suppress the noise, the noise of the image can be suppressed without compromising the spatial resolution. As a result, an accurate region extraction process can be performed.

In addition, an embodiment similar to this has already been described in the above-described case of `` comprising the pixel value v (x) with one image '' (see FIG. 14).

According to the present invention, the noise can be sufficiently suppressed without causing the blurring of the image, and can be effectively contributed to other image processing techniques such as pattern recognition techniques.

Claims (9)

Means for determining a similarity between the first pixel of the first image and each of the plurality of second pixels around the pixel corresponding to the first pixel on the second image; The weighted average value of the value of the first pixel and the value of the plurality of second pixels is calculated according to weighting values having a relatively high value when the similarity is relatively high and a relatively low value when the similarity is relatively low. Means An image processing apparatus comprising: Digitizing means for digitizing a weight having a relatively high value when the similarity is relatively high according to the similarity between each of the first pixel and the plurality of second pixels surrounding the first pixel, and having a relatively low value when the similarity is relatively low; An average means for weighted average of the values of the first pixel and the plurality of second pixels using the numerical weights; An image processing apparatus comprising: The method according to claim 1 or 2, And said averaging means obtains a new pixel value for said first pixel by weighted average. The method of claim 1, And said averaging means comprises determining means for determining a weighting coefficient based on said determination result, and multiplication means for multiplying said first pixel value and said second pixel value by a weighting coefficient. The method of claim 1, The multiplying means multiplies the second pixel value by a large weighting factor when the similarity between the first pixel and the second pixel is high by the multiplication means, and when the similarity is low, the small weighting factor by the second pixel value. Multiply by And a weighted average of the weighted first and second pixel values. The method of claim 1, And the determining means determines the similarity degree between the first pixel and the second pixel based on a plurality of images obtained by photographing the same object at different times using an imaging device. The method according to claim 1 or 6, And the determining means determines the similarity degree between the first pixel and the second pixel based on a plurality of images obtained by photographing the same object under different shooting conditions using an imaging device. The method of claim 1, And the determining means judges the similarity degree between the first pixel and the second pixel based on a plurality of images obtained by photographing the same object using an imaging apparatus under different processing conditions. Means for determining a degree of similarity between the first pixel of one image obtained by photographing an object using an imaging device, and each of a plurality of second pixels around the first pixel on the image; The weighted average value of the value of the first pixel and the value of the plurality of second pixels is calculated according to weighting values having a relatively high value when the similarity is relatively high and a relatively low value when the similarity is relatively low. Means An image processing apparatus comprising:

Images