JPS63278183A - Picture distortion correction device - Google Patents

Abstract

PURPOSE:To properly correct a picture having large deviation or strain and to obtain a picture little in the strain by using a manual mode and an automatic mode successively to correct the strain. CONSTITUTION:A strain calculating part 8 fetches respective first and second measured pictures stored in memories 3, 4 to calculate the strain of the picture. A strain correcting part 9 corrects the strain of the second measured picture based on a strain vector calculated in the calculating part 8. A difference computing element 5 takes the difference the difference between the first measured picture and the second measured picture in which the strain is corrected. This differential output is a moving state function picture having no strain. When the strain or the quantity of the deviation between the picture is large and the correction cannot be attained, as a first step, a position is manually correct and in a second step, an automatic alignment is executed based on the quantity of the deviation. Namely, in the manual mode, from the pictures of the memories 3, 4, a simple difference is obtained by a differential device 5 to display the result on a CRT 7, the picture of the memory 4 is moved manually while viewing it so as to have minimum arch fact in the display picture and have the quantity of the deviation capable of alignment and sent to the corresponding part 9.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an apparatus for digitizing X-ray film, and particularly to a suitable apparatus for correcting image distortion during inter-image calculation.

[Conventional technology]

Conventionally, during inter-image calculations, the amount of deviation of each corresponding part of two images is calculated using the cross-correlation of that part, and based on this result, the distortion of one image is made to match the distortion of the other. A method for performing the correction is proposed in Japanese Patent Laid-Open No. 59-52359. Furthermore, it is determined whether the corrected image subjected to distortion correction is appropriate based on the amount of deviation obtained by cross-correlation, and if it is not appropriate, the corrected image is replaced with the corresponding one of the two images. A method has also been proposed in which the size of each section for cross-correlation is changed after the substitution, and the deviation amount is calculated again.

[Problem that the invention seeks to solve]

When the conventional apparatus cuts out an image into sections and performs Fourier transform, it performs window processing to ensure continuity of the image at the periphery of the sections. However, when this window processing is performed, the maximum amount of positional deviation that can be detected due to the influence of the window width is up to n/4 for the number n of pixels on one side of the cutout section. Therefore, when there is a positional deviation of n/4 or more, distortion correction cannot be performed due to a large number of errors.

Therefore, it is possible to increase the size of the cutout section, but although this has the advantage of increasing the maximum positional deviation that can be detected,
On the other hand, it has the problem that fine distortions cannot be detected. In other conventional devices, the problem with the above device is solved by changing the size of the cutting section to blow-in, detecting distortion for each change section, and performing correction based on it, and repeating this process to change the size of the deviation from a large deviation to a small deviation. Distortion correction can be performed up to. It has the advantage of However, in the X-ray film image data targeted by this device, the elements of positional deviation include patient movement, the deviation of the film itself during imaging, and the positional deviation of the film during digitization, and the maximum amount of deviation is Not only does it contain larger errors than DSA, but in order to use it for a huge amount of X-ray film image data with high resolution, the size of the cutout section must be made even larger and this process is repeated until the amount of distortion converges. Must be repeated. This requires the memory capacity of the distortion calculation calculation section to be enormous. Also, by repeating the interpolation process,
This has the disadvantage that fine data parts of the image are degraded.

SUMMARY OF THE INVENTION An object of the present invention is to provide an image distortion correction device for correcting image-to-image calculations for images with a large amount of deviation.

[Means for solving problems]

The above object is achieved by, as a first step, correcting a large amount of deviation without performing interpolation processing while viewing the image using an animal operation, and then performing automatic correction based on the amount of deviation corrected here.

[Effect]

According to the present invention, image deterioration due to interpolation processing is reduced,
In addition, inter-image calculations can be performed with a small capacity distortion calculation memory.

〔Example〕

An embodiment of the present invention will be described below with reference to FIG.

The image data acquisition system 1 is a digitizer that digitizes X-ray film, and the external storage device 2 is a magnetic disk.
Data that has already been digitized is stored on optical disks, magnetic tapes, etc. Memory 3 stores the first digital image, and memory 4 stores the second digital image. The first digital image and the second digital image form images for subtraction. Subtraction is as follows. Measure the measurement site to obtain a first image, further inject a drug (contrast agent) that increases the absorption of X-rays, and measure the measurement site at a time different from the time at which the first measurement image was obtained. Measure the same area as above and view the second measurement image. Next, the difference between the first image and the second image is calculated. This difference becomes an image that gives the function of the organ of interest, and this is a subtraction processed image. 1st. The second digital image is typically stored in memory;
The memory 3 will store the first digital image and the memory 4 will store the second digital image.

The image processing unit 100 takes in the image in the memory 3 and the image in the memory 4, and performs a difference calculation between the two. The result of this difference calculation indicates the amount of change at the measurement site. Here, the image processing unit 100
has a distortion calculation unit 8 that calculates distortion, and a distortion correction unit 9 that corrects it.

Next, let's talk about distortion. If the measurement site is an organ that moves, the movement information will be superimposed on the target functional information (difference information), making it impossible to accurately obtain the original dynamic functional image.

If only the movement information can be deleted, only the original dynamic mechanistic image can be obtained.

Therefore, we introduced the idea of distortion to remove only the movement information. This distortion refers to a factor that indicates information on the mutual movement between two measurement images that are the subject of a difference. This distortion is the first. Calculations are performed based on the second measurement image, and then the second measurement image is corrected based on the amount of distortion that is the calculation result, and movement information is deleted. Next, the difference between the first measurement image and the corrected second measurement image is calculated. The difference information thus obtained becomes the original functional image to be sought.

The memory 6 stores a functional image of the image processing section 100. CR
T7 displays a functional image of the memory 6. This display image is
It forms an image of the original function and serves as diagnostic information.

The image processing unit 100 includes a distortion calculation unit 8 and a distortion correction calculation unit 9.
, and a difference calculating section 5. The distortion calculation unit 8 is connected to the memories 3 and 4.
import the first measurement image and second selected image stored in the
Calculate image distortion. Calculation of image distortion is performed in the first step. Second
This refers to cutting out a measurement image into small sections, performing cross-correlation calculations between corresponding sections, and calculating a distortion vector indicating distortion for each section.

The distortion correction calculation unit 9 performs distortion correction on the second measurement image based on the distortion vector calculated by the distortion calculation unit 8.

The difference calculation unit 5 calculates the difference between the first measurement image in the memory 3 and the second measurement image after distortion correction. This differential output is
This results in a dynamic and functional image without distortion.

On the other hand, the feature of this embodiment is that when the amount of distortion or deviation between images is large and cannot be corrected, manual position correction is performed as a first step, and the second step is performed based on this amount of deviation.
The point is that automatic alignment is performed in steps. Here, the manual positioning is performed by the operator by looking at the displayed image on the CRT 7. In other words, manual mode means that the images in memory 3 and memory 4 are simply subtracted by subtractor 5, the result is displayed on CRT 7, and the image in memory 4 is input externally so that artifacts in the displayed image are minimized. This is a moving mode, and this manual alignment adjusts the amount of distortion to the amount of deviation that allows automatic alignment. Next, the amount of deviation determined in this mode is sent to the distortion correction section 9, and is used as a corrected live image for automatic alignment based on it.

An embodiment of the distortion calculating section 8 is shown in FIG. The distortion calculation unit 8 includes: (i) a division extraction processing unit 11 using a sample window;
12, (3i) FFT calculation units 13, 14 that perform Fourier transformation on the cutout output of the division and extraction processing unit; (iii) Filter region unit (space) that performs filter region processing on the complex Fourier space on the output of the FFT calculation unit; filter) 15, 16, (iv) memory 3 and 4
A cross-correlation calculation unit in Fourier space that performs mutual processing with the outputs of the spatial filter regions obtained for each image, that is, a complex conjugate calculation unit] 7. (v) An inverse Fourier transform of the output of the cross-correlation calculation unit It consists of a transformation section 18 and (vl) a shift amount calculation section 19 that calculates the shift amount from the output of the inverse Fourier transform section.

(i) Division cutout processing unit 11 using the sample window
, 1.2 is shown in FIG. The processing contents in the processing units 11 and 12 are the same, and the only difference is whether the processing target is the image in the memory 3 or the image in the memory 4 that has undergone rough misalignment correction in the manual mode.

Therefore, the processing contents in the processing section 12 are shown below.

In FIG. 3, the original image refers to data stored in memory.

The original image can be displayed in two-dimensional coordinates, R(x.

y). Original image R (x r y), vertical size m,
It is cut out as one section of m x n size (meaning m pixels x n image) with horizontal size n. This section is cut out in the order from upper left to upper right and from lower left to lower left. This cutting order follows the rask scan. Therefore, by cutting out, a plurality of sections are cut out from one original image.

This cutting section is fixed in the method disclosed in Japanese Patent Laid-Open No. 59-52359. In this embodiment, as described above, the size of the section is changed depending on the occurrence of artifact 1-. Even if the section is changed in this embodiment, the specified section remains the same at each time of the change. The general idea regarding the original classification of the specified classification is described below.

Each section has a center region and a region near the border. The area near the boundary is near the boundary of the partition, and the partitions separated by the boundary are treated independently, so there is strong discontinuity as data, and the central area of the area is far from the partition boundary, so it is difficult to use as data. Strong continuity.

Therefore, in order to remove the adverse effects of discontinuities outside the central region due to image cutting, it is necessary to emphasize the central region and compress the regions near the boundaries.

Therefore, we set the extraction window function W(j,,j),
A cutting window function W (ITJ) is set and integrated with the cutting section. The cutout window function W(i,j) is a function that performs center emphasis and boundary area compression, as shown in FIG. The cutout image Q, (ko, j
) is Qpq(i+j)=W(Lj)XR(i+mp,j
+nq) ・a). Here, p indicates a cutout section number viewed in the horizontal direction, and q indicates a cutout section number viewed in the vertical direction, and the section number is specified by the combination of p and q.

Therefore, the meaning of R(i +mp, j +nq) is p
An image R within a section that can be specified by and q is shown.

The cropping window function W(i, j) is an error due to the effect of image cropping in post-processing (truncation error).
In order to reduce this, it is essential that the side lobe be small and that it be a finite function.

Now, if we express the cutout wind function W(i, j) in polar coordinates, W(i, j)=G (r)...
・(2) becomes. If the origin of the polar coordinates is set at the center position of each section, r becomes (]1).

Furthermore, when G(r) is given by Blackman's function (the maximum side lobe is -4.0 dB), the following is obtained. When given by trigonometric functions, it becomes . However, the maximum sidelobe in this case is -28d
It is B. The definitions and roles of the above-mentioned cutout wind function W, truncation error, side lobe, finite function, Blackman's function, etc. are all well known. Publicly known documents include: ■ "Digital Signal Processing" by Miyayo et al., Institute of Electronics and Communication Engineers, Corona Publishing, published in 1973, ■ B1ackm
an, R, B and J, W, Tukey r.
The measurement of powerS
pectral New York, Dover, l
It is 985.

The calculation of equation (1) in the above extraction process 1]- is as follows. A cutting window function is stored in a memory different from the memory 3 for each coordinate (i, j) of the cutting section.

The window function of this memory and the data corresponding to the partitions of the memory 3 are integrated for each coordinate unit. This integration result becomes the calculation result of equation (1).

The integration results are temporarily stored in a buffer memory and prepared for the next processing.

In FIG. 4(a), the horizontal axis represents the frequency ω and the frequency spectrum F(ω) represents the vertical axis. The display unit of frequency ω on the horizontal axis is
LP/cm. Here, LP means line pair. In the figure, Tikis 1 is displayed on the horizontal axis.
~Frequency ωN means the maximum frequency that can be expressed.

In this embodiment, it refers to the maximum frequency that can be expressed by the extracted pixel space.

Figure 4 (b) shows the frequency spectrum F(ω) as F
□, F2. We have disclosed three patents of Fs, each with its own characteristics. Furthermore, the magnitude of the frequency ω is roughly divided into three, and these are E, E, and E2. I set it as E 3.

Region E1 is a region for reducing the effect of the extraction window, region E2 is a region for noise removal, and region E3 is a region for emphasizing the spatial frequency component of interest (for example, a region of 1.0 to 0.2 LP/σ). The characteristic G(ω) of the dashed-dotted line in the region El is the frequency spectrum due to the cut-out wind function W. Ideally, the characteristics of the spatial filter should be set to zero in order to reduce this frequency spectrum as much as possible. Set.

Characteristic F1 is the most commonly adopted characteristic and is in the area E.
In the machining process, the setting was made to reduce the influence of the cutting wind. That is, the frequency spectrum G(ω) of the cutting window ≠
In a region Eo that is not O, the frequency spectrum of the characteristic F1 is set to O. Characteristic F2. F3 has similar characteristics in area Eo
I let him have it.

The characteristic F1 has a peak in the region E3, and in the region E2, the characteristic gradually decreases to O at the Nyquist frequency. Therefore, when a spatial filter has characteristic F1, it is possible to remove the negative effects caused by extraction (area El), and in the area of the spatial frequency of interest, that frequency can be emphasized (area E3), and it is possible to eliminate high frequency noise. It was possible to reduce the level and reduce the noise (
Area E2).

The characteristic F2 has a characteristic in which the spectrum peak has shifted to the right compared to the characteristic F1, and the characteristic F3 has a characteristic in which the spectrum peak has shifted to the left compared to the characteristic F1.

The characteristic F2 has a frequency spectrum larger than that of F'' in the region E2. Therefore, it is adopted as a characteristic of a spatial filter for waveforms that are high in frequency and have few noise components.

The characteristic F3 is a characteristic suitable when there are many high frequency noise components, contrary to the characteristic F2.

Which of the characteristics F11, F21, and F3 to adopt depends on the nature of the waveform to be analyzed. In addition to this characteristic, there are various other characteristics. Although the peaks of the characteristics Fz and Fil Fs are all the same, they may be different, and in the region E 11 E
2J E s can also be set arbitrarily.

The characteristics and effects of the above spatial filter are well known. References include rDigital Ficoters
JR, W, Hamming, Be1l Labor
[atories and Nava]postgrad
uat 5choo1. 1977, Psenti
ce-Hall Inc. This document also describes the above-mentioned wind function.

The only difference is that the processing target of the cutout processing unit 12 is the corrected image of the memory 4, and the other processing is completely the same. The same function value may be set for the extraction window function W(i, j).

The extraction processing units 11 and 12 extract and classify the read data from the memories 3 and 4 according to which category the address belongs to. Therefore, when changing the classification, it is only necessary to change the contents of the address management to change the classification.

[...]'s FFT calculation units 13 and 14 take in the outputs of the extraction processing units 11 and 12, respectively, and perform FFT (fast Fourier transform) calculations. This results in a conversion to the frequency domain.

The spatial filters 15, 1.6 in FIG. 2 remove the influence of the cutout wind function, emphasize the alignment element (spatial frequency component of interest), and remove noise.

This improves the sensitivity of partial correlation in the next process.

The spatial filters 15 and 16 perform spatial filter region not on real space but on Fourier space. Spatial filter region in "Fourier space" means to perform filter region in the frequency domain.

FIG. 4 shows a characteristic diagram of the spatial filters 15, 1.6, and FIG. 5 shows a detailed embodiment diagram centered on the spatial filters 15.16.

Figure 4 (b) shows the desired frequency characteristics of Figure 4 (a)
It is a figure which shows the application method when applying to a dimensional image. That is, assuming that the center point 0 in Fig. 4 (b) is the zero frequency and the right end (+ r Pax ) is the Nyquist frequency, each 2
The characteristics shown in FIG. 4(A) were expressed by setting the distance from a point on the dimensional coordinates to point 0 as the frequency r. Here, point 0 is the position of zero frequency where the frequency ω in Fig. 4 (a) becomes ω = 0,
That is, it means the same point as the origin 0.

The above spatial filter region is performed for each cutout section. The spatial filter region is an operation that multiplies the Fourier transform image of each section by a filter function. On this occasion,
Since the filter functions to be integrated are only real numbers (the imaginary part is zero) and the Fourier transform images are both expressed as complex numbers,
The product operation becomes the real part of the image after Fourier transformation. The results of this product operation are stored in the memories 24 and 25 as complex numbers.

Next, the configuration and operation of FIG. 5 will be explained. Spatial filter 1
5.16, multipliers 21, 22, address generator 23.
23A, memory 24. ．． 25, and a memory 2o. Memory 20 and address generator 23 are connected to spatial filter 15.
16 are common to each other. The multiplier 21 multiplies the output of the FFT calculation unit 13 and the output of the memory 20 by a complex number.

The multiplier 22 combines the output of the FFT calculation unit]-3 and the memory 20.
Multiply with the output of . These two multiplications involve multiplying the frequency spectrum, which is the result of the FFT operation, by the spatial frequency characteristics shown in Figure 4 (a) in the complex Fourier space, thereby obtaining the frequency spectrum that reflects these characteristics. obtain. The frequency spectrum that reflects the characteristics of FIG. 4 (a) is one in which the cutout wind frequency is reduced, that is, the bad influence of the cutout window is removed, and noise is removed.
It has the following characteristics.

The memories 24, 25 store the multiplication results of the multipliers 2], 22 under instructions from the address generators 23, 23A.

This storage is performed in coordinate units of divided sections. On the other hand, during the next stage of cross-correlation, the address generators 23 and 23A perform rask scan-like addressing of the divided sections for the memories 24 and 25. Each of the memories 24 and 25 has two memories for storing a real component and an imaginary component.

The cross-correlation calculation unit 17 includes a calculation unit 26 and a memory 27. The arithmetic unit 26 performs a complex conjugate operation on one of the data stored in the memory 24 and the data stored in the memory 25, and multiplies the data by the other. Here, complex conjugate operation means to convert a complex number into a+bi
When , we are referring to the operation a-bi. However, j indicates an imaginary number.

Cross-correlation in Fourier space 2 means performing a FIXF2 calculation, assuming that Fl and F2 are functions that should be correlated with each other. This is a known fact. Obtaining F2 from Fl is a complex conjugate operation, and calculating FIXF2 is a multiplication. Therefore, taking cross-correlation in Fourier space means performing a complex conjugate product operation.

The reason for taking the cross-correlation between two functions is to quantitatively find out how similar the two functions are to each other.

The cross-correlation calculation results are stored in the memory 27.

This cross-correlation calculation is performed for each segment that has been cut out.

The inverse Fourier transform unit 18 performs inverse Fourier transform on the complex conjugate product operation result, which is the cross-correlation result stored in the memory 27 . As a result, a cross-correlation calculation result in real space is obtained. For example, assume that the cutout section is mXn. However, m and n are vertical and horizontal cutout 1], and m and n generally mean the number of cut out pixels. - As an example, m=8. n=8 or m=16, n-16, or m=32. It has a cutting width such as n=32. m≠n may also be satisfied. In cutting out the division of mXn, the cross-correlation is performed in units of its intermediate coordinates,
mXn cross-correlation values are obtained in one section.

Note that cross-correlation can be performed not only in Fourier space but also in real space.

Next, as shown in FIG. 2, the output of the inverse Fourier transform section 18 is taken into the deviation amount calculation section 19.

The configuration of this deviation amount calculating section 19 is shown in FIG. The deviation amount calculation unit 19 includes a peak detection unit 30, a memory 31, and an interpolation unit 3.
2. It consists of a memory 33.

The peak detection unit 30 detects a peak value (maximum value) from m×n cross-correlation values obtained for each section. If the number of divisions is M, then M peak values can be detected for one image. The peak value within one section indicates that the functions that take the cross-correlation within that section are most similar.

The coordinate on the two-dimensional coordinate that gives this peak can be considered to indicate the direction of deviation. Therefore, in the present invention, the coordinates on the two-dimensional coordinates when giving this peak are defined as strain vectors, and this strain vector is determined for each section.

The memory 31 stores the distortion vector detected by the peak detection section 30. The interpolation calculation unit 32 calculates each distortion vector of all pixels from the distortion vector of each section by interpolation. Each distortion vector of all pixels is the first. This is the distortion vector I for each pixel between the second measurement images. The memory 33 stores the distortion vectors l~ determined for all the pixels.

FIG. 7 is an explanatory diagram of distortion vectors in division units, and FIG. 8 is an explanatory diagram of distortion coordinates. In FIG. 7, memory 24 stores the spatial filter output for image A.

It is assumed that both images A and B are cut out into 5×5 sections.

Furthermore, it is assumed that one section is cut out by 5 (pixels) x 5 (pixels). Memory 25 stores the spatial filter output for image B. The size of one divided section of image A is 5
When ×5, At to A55 indicates the spatial filter output of one division. By setting the size of one division of image B to 5×5, 11311 to 855 indicate the spatial filter outputs of one division.

The processing unit 40 includes the cross-correlation calculation unit 26 to the peak detection unit 3
This is a general term for processing systems that reach 0.

The memory 31 corresponds to the size of one image and is composed of 5×5 sections. The correlation between the sections is the spatial filter output A
This is a cross-correlation between 1s~A55 and B11~Bri5. A
What is necessary is to take the cross-correlation between I and Bll, the cross-correlation between A12 and 812, and the cross-correlation between A55 and 85F+.

The results of the complex conjugate and product operations between the sections are subjected to inverse Fourier transform to obtain respective cross-correlation values. The cross-correlation value is determined for each pixel point in one section.

Therefore, if the segment is a combination of 5 x 5 pixels, 5 x 5 cross-correlation values are obtained for each segment. The maximum value among these 5×5 cross-correlation values becomes the peak value in that section.

The coordinates in the two-dimensional image when providing the peak values for each section become distortion vectors and are stored in the memory 31. 611
~C51) is a distortion vector in each section. This distortion vector represents the center coordinates within the section as the origin. For example, C11 indicates the strain vector in the first section, and the arrow pointing diagonally to the right indicates a vector connecting the coordinate point where the peak value was obtained and the central origin. As is clear from the figure, it is possible to obtain a distortion vector for each section, and to know in which direction the image is moving for each section.

FIG. 8 shows the relationship between memory 31 and memory 33. The memory 31 stores distortion vectors C in units of sections. The memory 33 indicates distortion coordinates N obtained by interpolating pixel by pixel from the partition using this distortion vector. The distortion coordinates in FIG. 8 show only a typical line diagram, but in reality, the line diagram is constructed pixel by pixel by the interpolation calculator 32, and is composed of denser coordinate axes.

Returning to the explanation of FIG. The deviation amount calculation unit 9 includes an interpolation calculation unit (
It consists of a coordinate conversion section) 34, a memory 35, and an adder 36 that adds the amount of deviation determined in the manual mode of the first step to the distortion vector coordinates in the memory 33. Coordinate conversion section 34
corrects the pixels in the memory 4 using the distortion coordinates in pixel units in the memory 33 obtained by automatic alignment and the added distortion coordinates of the shift of t−i for the entire image obtained manually. This coordinate transformation is a correction to match the distortion of the image in the memory 4 to the distortion of the image in the other memory 3, so that the distortions of the two images match. The image data of the memory 4 after correction is stored in the memory 35.

The difference calculation unit 5 calculates the difference between the images in the memory 35 and the memory 3. The result of this difference is that images A and B are free from distortion. This results in a dynamic functional image without distortion as a result of the difference.

Next, as a second embodiment, it is also possible to use the first step automatic alignment instead of the manual mode correction described above. That is, in the first embodiment, manual and automatic alignment was used, whereas in the second embodiment, automatic alignment was performed using an automatic aligner. In other words, in the second embodiment, the ninth
As shown in the figure, the size of the 7 links used for automatic alignment is set large for the first alignment, and the rough peak deviation between images is determined and this is stored in the memory without interpolation.
0, and based on this, the same automatic alignment as in the first embodiment is used. In this way, in the [th]-th alignment, a large amount of deviation is obtained and the second
By performing automatic alignment, it is possible to reduce errors caused by interpolation and obtain a large amount of deviation.

The second embodiment below is suitable for obtaining a dynamic functional image by digitizing X-ray film and subtracting the images, and is particularly suitable for subtracting continuously captured images using an X-ray film auto changer. There is. Furthermore, this embodiment can be applied to fields other than medicine.

Inter-image operations include operations that take the difference between two images, 2
There are operations such as calculating the product of two images, calculating the ratio (division) of two images, and calculating the sum of two images. Although the present invention has shown only difference calculations, it can be applied to all of the above types of calculations. The difference between these, as shown in FIG. 8, is that the difference calculation section 4 is replaced by another calculation section. Furthermore, inter-image calculations can be used in the field of robot control and pattern recognition in addition to the field of ME.

〔Effect of the invention〕

According to the present invention, since distortion correction is performed continuously in manual mode and auto mode, it is possible to perform appropriate correction even for images with large deviations or distortions such as X-ray film images, and to produce images with little distortion. Now you can get it.

[Brief explanation of drawings]

FIG. 1 is a diagram for explaining an embodiment of the device of the present invention, FIG. 2 is a diagram for explaining the distortion calculation section of FIG. A) and (B) are diagrams explaining spatial filtering, FIG. 5 is a diagram explaining the spatial filter of FIG. 2, FIG. FIG. 7 is a diagram for explaining spatial filter data and intra-section distortion vectors, and FIG. 8 is a diagram for explaining intra-section distortion vectors and distortion coordinates interpolated for each pixel. FIG. 9 is a diagram illustrating the distortion vector when the 7-trix is increased. 1...Data detection system, 2...External storage device, 3, 4...
・Memory, 5 Difference calculator, 6 ・Display memory, 7 ・CRT, 8... Distortion calculation unit, 9 ・Distortion correction unit,
10...Manual correction section, 11.12...Cutout processing section, 13.14...FFT calculation section, 15.16.
... Spatial filter, 17... Cross-correlation calculation unit, 18.
・・Inverse Fourier transform unit, 19 ・・Shift amount calculation unit, 20
゜24.25.27...Memory, 21.22...Integrator, 23.23A...Address generator, 26...
Arithmetic unit, 30...Peak detection section, 31, 33.35.
...Memory, Figure 4 (4 things (b))

Claims (1)

[Claims] 1. During inter-image calculation, the first means corrects the rough deviation using parameters from external input while checking the image deviation, and the second means corrects the deviation based on the amount of deviation obtained by the first means.
A second means of calculating the amount of deviation of fine parts using the cross-correlation of each corresponding part of two images, and based on this result, matching the distortion of one image to the distortion of the other image. An image distortion correction device for image-to-image calculation, comprising automatic distortion correction means.