JP3091935B2 - Multi-value estimation method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multi-value estimation method for estimating and outputting a multi-value image from a binary image.

[0002]

2. Description of the Related Art When handling two-dimensional image information, generally, an image is finely divided into small regions called pixels. For each pixel,
There are cases where the information is handled by binary information of white or black, and cases where it is handled by multi-valued information having a plurality of stages by finely separating white and black. Although it is more preferable to use multi-valued information to obtain a more natural image and to improve the image quality, there is a disadvantage that the amount of information of the image becomes large and is not suitable for recording or transmission.

On the other hand, when handling with binary information, the information amount of the image is smaller than that of the multi-valued information and is suitable for recording and transmission, but the image quality is not always sufficient. A pseudo-halftone binarization method such as a dither method that represents multi-values by devising the arrangement of binary pixels is also used, but this method also has poor image quality when compared with a multi-value image in which each pixel has multi-value information. I can not deny that.

If multi-valued information can be estimated from a binary image, it is handled in the form of binary information at the time of recording and transmission, and is used at the time of displaying on a display or printing out on paper. By using the information by estimating the multi-valued information, the advantages of both the binary image and the multi-valued image can be obtained.

There have been several attempts to restore a binary image to a multi-valued image. Among them, the most common method is a method called a window method. Image restoration technology for halftone images ”Masahiko Matsunawa / Konica Corporation /“ Image Lab ”1990.3:
Reference 1] and the like.

In the window method, a reference range, that is, a window is set on a binary image around a target pixel position to be restored.
The estimated pixel value of the target pixel position is determined based on the ratio of black / white pixels in the window.

FIG. 6 shows an example of a window used in a conventional window system. FIG. 6A shows a window for a halftone portion which emphasizes tone reproducibility, and FIG. ) Is a window for a character part which emphasizes the preservation of edge components. The numerical values in a number of sections in each window indicate the weighting coefficient of the area. In the halftone window [see FIG. The part window [see FIG. 7B] has a window configuration in which only the center pixel (pixel of interest) is weighted.

In this window method, when the size of the window is increased, the number of pixels to be referenced is increased and the gradation reproducibility is improved, but edge components such as characters and figures are blurred. Conversely, when the window is made smaller, the preservation of edge components such as characters and figures is improved, but the gradation reproducibility is reduced because the number of reference pixels is reduced. As a countermeasure for this, it is common to weight the influence (each numerical value in FIG. 6) according to the pixel position in the window.
The gradation reproducibility and the edge component preservation can be changed depending on the size of the window and the way of assigning the weighting coefficients.

As a general tendency, it can be said that gradation reproducibility and edge component preservation have a trade-off relationship. That is, if the tone reproducibility is emphasized, the edge preservability is not sufficient, and if the edge preservation is emphasized, the tone reproducibility is not sufficient. For this reason, when the estimation is performed for a single image with a single window, sufficient image quality may not be obtained. This is because the image characteristics are different depending on each region even in one same image. For example, in the case of an image such as a resume in which a character and a person image are mixed, edge preservation is important in a character part, and gradation reproducibility is important in a halftone part such as a person image.

As described above, since the characteristics vary depending on the region even in one image, in order to perform a process suitable for the characteristics,
Conventionally, it has been proposed to divide an image region into pixels or blocks and restore each region by a different method. The classification based on the characteristics referred to here is, for example, dividing into a character area and a halftone area according to the strength of an edge component. Restoring by another method means, for example, that a reference window used for multi-value estimation is used in a character area. It is to make it small and make it large in the halftone area.

The processing method using this area discrimination is described in ["Recovery method from binary image to multi-valued image using area discrimination" Shigenobu Fukushima / Minolta Camera Co., Ltd.
/ 1991 Image Electronics Society Annual Conference: Reference 2]. Examples of such a method include a method of selecting an optimal window size and shape from image characteristics at a target pixel position of a target image (references 1 and 2)) and a method of once performing multi-value restoration and then performing binary conversion again. And a method of selecting a restoration method so as to match the original binary image [Publication of Japanese Patent Application Laid-Open No. 62-295567
"A method of estimating a halftone image of a binary image", Seiichiro Hiratsuka, et al./Konishiroku Photo Industry Co., Ltd .: Reference 3], etc., have been proposed.

However, it is difficult to divide an area, and it is not always possible to obtain a desired result of division.
If there is an error in the classification, it may have a great influence on the image quality, and may worsen the impression of the entire estimated multi-valued image.

The advantage of performing the multi-value restoration is that, in addition to the improvement of the image quality of the restored image by the multi-value restoration, the reprocessing suitability is improved. For example, when reprocessing such as reduction or enlargement is applied to a binary image binarized by the tissue dither method,
Interference may occur due to the periodicity caused by the tissue dither matrix, resulting in an unsightly image. In such a case, the binary image can be multi-valued to reduce the periodicity of the dither matrix, thereby preventing image quality deterioration.

[0014]

As described above, in the above-described conventional multi-value estimation method, since the characteristics are different for each region even in one image, uniform multi-value restoration is performed on the entire image without dividing the region. In this case, there is no choice but to sacrifice either the resolution or the tone reproducibility, or to take a mediocre balance in both. Therefore, it was conceived to apply a multi-level restoration method using different windows for each area by dividing the area. In this case, it is possible to achieve both resolution and gradation reproducibility, Therefore, there is a problem that the image quality tends to be remarkably deteriorated at the position where the classification error has occurred.

The present invention eliminates the above-mentioned problems, and performs a multi-value restoration process suitable for each region by dividing the region. It is an object of the present invention to provide a multi-value estimation method that suppresses and contributes to improving the quality of the entire estimated multi-value image.

[0016]

SUMMARY OF THE INVENTION The present invention provides a multi-value estimation method for estimating and outputting a multi-value image from a binary image having binary information of whether each of the pixels constituting the image is white or black. An image characteristic determining unit that refers to an array of each pixel in a peripheral region for each pixel constituting a binary image and determines an image characteristic of each pixel based on the array, and attempts to perform multi-value estimation. A multi-valued image estimating means for setting a reference region around the pixel of interest and estimating a multi-valued image of the pixel of interest based on a pixel value and a weighting coefficient of each pixel in the reference region; A density diffusion unit for distributing a weighting coefficient of each pixel in the reference area according to the image characteristic of each pixel in the reference area including the target pixel.

[0017]

[0018]

According to the present invention, when estimating the multi-valued image of the pixel of interest based on the pixel value and the weighting coefficient of each pixel in the reference area set around the pixel of interest to be multi-valued, The weighting coefficient of each pixel in the reference area is distributed according to the image characteristics of each pixel in the reference area including the pixel. According to this processing, when estimating the multi-valued image, the determination result of the image characteristics is referred to at a plurality of pixel positions in the reference area. Therefore, even when erroneous determination of the image characteristics occurs at one pixel position, the estimation is performed. The influence on the density of each pixel of the multi-valued image can be reduced, and as a result, the quality of the estimated multi-valued image can be improved.

[0019]

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below in detail with reference to the accompanying drawings. FIG. 1 is a conceptual diagram of a multi-value estimation method according to an embodiment of the present invention. In FIG. 1, reference numeral 101 denotes an input binary image; 102, an image characteristic determining unit that determines an image area for each pixel position based on characteristics such as pixel arrangement of the input binary image; A density diffusion unit 104 that receives the result and diffuses the density of the pixels of the input binary image 101 into the output image, and 104 is an estimated multi-value for the input binary image output after receiving the density diffusion processing by the density diffusion unit 103. It is an image.

Hereinafter, the operation of the multi-value estimation processing according to an embodiment of the present invention will be described in detail with reference to FIG. First, the input binary image 101 has pixels arranged in a two-dimensional direction, and each pixel has a binary value of white or black. In particular, in this example, it is assumed that a pixel value of 0 indicates a white image and a pixel value of 1 indicates a black image.

On the other hand, with respect to the output estimated multi-valued image 104, the arrangement of pixels is assumed to be the same as the arrangement of the binary image, and each pixel has a plurality of values between white and black. I do. In this example, it is assumed that the pixel value takes a positive integer value in the range of 0 to 255, 0 is white, 255 is black, and the value between them changes sequentially between white and black. Thereby, for example, the pixel value 128 becomes a gray image.

In FIG. 1, an input binary image 101 is supplied to an image characteristic discriminating section 102 and a density diffusion section 103. The image characteristic determination unit 102 determines an image area for each pixel position of the input binary image 101, and outputs a result of the determination to the density diffusion unit 103. In this determination, the image area is classified into, for example, a character area and a halftone area such as a person's face, and the arrangement of the black / white pixels and the change in the distribution density of the black / white pixels in the binary image. For example, the image area can be determined from the above. Also, it is of course possible to determine the area according to the instruction of the operator.

FIG. 2 is a conceptual diagram for explaining the density diffusion processing in the density diffusion section 103. Nine areas separated by straight lines indicate independent pixels. Here, the operation when the center pixel painted black is processed will be described.

First, the characteristics of an image at this position are obtained from the ratio of white / black pixels and the distribution shape of binary pixels in the pixel and its surrounding area. For example, when a halftone image portion such as a person's face is binarized by error diffusion, pixels are likely to be isolated. By examining the characteristics of a reference region around a target pixel position, the pixel characteristics are determined for each pixel position. It is possible to determine.

Next, in accordance with the result of this determination, the density of the target pixel is distributed to the target pixel itself and its surrounding area.
For example, in a pixel determined to be a character portion, diffusion is limited to the pixel of interest itself to prevent blurring of the image, and in a pixel determined to be a halftone portion, an estimated multi-valued image is diffused widely to the pixel of interest and its surrounding area. To make a gradual change. In FIG. 2, arrows indicate the state of the diffusion.
However, in the figure, diffusion to the target pixel position is omitted.

Next, the processing operation of the density diffusion section 103 will be described. In order to make the explanation easier to understand, here, the multi-value estimation method using the conventional window method will be described first. As described above, in the window method, a reference area (window) is set around a target pixel to be multi-valued, and a multi-valued estimated value is obtained from a ratio of black / white pixels of binary pixels in the reference area. Things. FIG. 6 shows an example of a reference area (window) used in the conventional window method, and the number described at each pixel position indicates a weighting coefficient. FIG.
Is a window used for a pixel determined to be a halftone portion,
They refer to the surrounding area widely. On the other hand, FIG. 2B shows a window used for a pixel determined to be a character portion, and refers only to the binary pixel value of the target pixel position itself. The fact that the weighting coefficient is 0 is equivalent to not actually referring.

First, the multi-value estimation processing for the halftone portion will be described. Now, assuming that three pixels in the window are black and the remaining pixels are white, (the number of black pixels in the reference area) / (the total number of reference pixels) becomes 3/9, and in the case of 255 gradations, 85 is multi-valued estimation. Value. Since the weighting coefficient of the window used here is constant, the description of the weighting calculation is omitted. On the other hand, in the multi-value estimation processing for the character portion, since the weighting factors other than the target pixel position itself are 0, the multi-value estimation value depends only on the binary pixel value at the target pixel position.
That is, if the target pixel is black, the multi-valued estimated value is 255, and if it is white, it is 0.

When the above multi-value calculation method is expressed by an equation, (multi-value estimation value) = (binary pixel value in reference area × weighting coefficient at each reference position) / (sum of weighting coefficients) ‥‥‥
The relationship (1) holds.

The weighting coefficient at each reference position differs depending on the window to be used, and the window to be used is determined based on the image characteristic determination result at the target pixel position. In the above equation (1), the case where the binary pixel value is 1 is black, and the case where the binary pixel value is 0 is white, and the multi-valued estimated value is a real number between 0 and 1. As described above, in order to obtain an integer between 0 and 255, the value may be multiplied by 255 to round a decimal point or less. In addition,
The shape of the reference area used by the image characteristic determination unit 102 and the shape of the reference area used by the density diffusion processing unit 103 may be completely different, but if they match, the storage area can be used effectively.

FIG. 3 is a conceptual diagram showing the density diffusion method of the present invention relating to the multi-value estimation processing in comparison with a conventional window method. In the figure, each area divided by a straight line indicates a pixel, and an operation when a central pixel is processed will be described. Here, the straight line with the arrow indicates the magnitude of the influence of the reference pixel in the window on the calculation of the multi-valued estimated value at the center position, and the thickness of the line represents the magnitude of the influence. ing.

In general, importance is placed on the preservation of edges in a character portion, so that the binary pixel value of the target pixel itself is emphasized, and the binary pixel values of peripheral pixels are not. On the other hand, in the halftone portion, since the gradation is important, the target pixel itself and its surrounding binary pixel values are widely referred to. In the conventional window method, as shown in FIGS. 3A and 3B, one of the windows is selected according to the image characteristic determination result at the target pixel position, and the selected window is used to select a binary pixel in the reference area. The magnitude of the effect of the value was determined.

On the other hand, in the density diffusion method according to the present invention, according to the result of image characteristic determination at each pixel position,
The magnitude of the effect that the value pixel value has on the surrounding pixel positions is determined. To compare this with the conventional method, FIG. 3 (c) changes the viewpoint.

From the viewpoint of the magnitude of the influence of the binary pixel value of each reference pixel upon multi-value estimation at the target pixel position, the magnitude of the influence of each reference binary pixel is determined at each pixel position. Since it is determined based on the image characteristic determination result, FIG.
As shown in (c), the size of each reference pixel has a different effect. Of course, if the image characteristic discrimination result at each pixel position in the reference region is single, the magnitude of the influence of each reference binary pixel is the same.
It goes without saying that this is the same as (b).

Next, advantages of the concentration diffusion method of the present invention will be described with reference to FIG. FIG. 4 shows an example of a case where only one pixel is determined to be a character portion in an area determined to be a halftone portion, that is, an erroneous determination of image characteristics occurs. In FIGS. 6A to 6D, pixels painted black are pixels that are determined to be character portions, that is, pixels that are erroneously determined.

FIGS. 7A and 7B show the processing of the conventional window system, and FIGS. 7C and 7D show the density diffusion system of the present invention. For the window method, a window having the weighting coefficient shown in FIG. 6 is used, and the same coefficient window is used for the degree of diffusion in the density diffusion method.

In the case of the conventional window method, the window is determined based on the result of image characteristic determination at the pixel position of interest (here, the center pixel position). Windows such as those shown in FIGS. 4 (a) and (b) are used. Now, assuming that black pixels and white pixels are distributed in approximately the same number in this area, FIG.
In the position A shown in FIG.
At the position B shown in FIG. 4B, the multi-valued estimated value becomes either black (255) or white (0) depending on the binary pixel value of the target pixel. This pixel becomes a singular point on the multi-valued estimated image, and becomes a major cause of deterioration in image quality. Also, if most of the pixels in this area are white pixels, and if the pixel of interest in FIG. 4B is a black pixel, it also becomes a singular point, which is a major factor in image quality degradation.

On the other hand, in the density diffusion method, since the image characteristics are determined for each reference pixel position and the magnitude of the influence, that is, the weighting coefficient is determined, the multi-value estimation of the pixels at the position A and the position B is performed. A window as shown in FIGS. 4C and 4D will be used. As is clear from the comparison of FIG. 6D with FIG. 6B, the influence of the reference pixel does not become zero even at the pixel position where the erroneous determination has occurred in the density diffusion method, so that the function of suppressing the deterioration in image quality is achieved. have.

FIG. 5 shows the results of subjectivity evaluation of processed images of the conventional window system and the density diffusion system of the present invention. The image to be evaluated includes the test chart No. for facsimile of the Institute of Image Electronics Engineers of Japan. An image obtained by binarizing a multi-valued image 1 by a scanner with an average density neighborhood method is used as an original image. Further, a face portion of a person and a portion of Japanese characters are cut out from this image, and a multi-valued image is obtained by applying window processing and density diffusion processing to each image. The image after the processing was displayed on a CRT, and evaluated on a five-point scale subjectively. The breakdown of the five stages
5: good 4: slightly good 3: normal 2: slightly bad
1: Bad. The evaluators are 10 image specialists. The discrimination error is randomly generated by a pseudo random function on a computer.

In FIG. 5, the abscissa represents the discrimination error rate as a percentage, and the ordinate represents the average of the evaluation points. FIG. 11A shows the evaluation result in the character portion, and FIG. 10B shows the evaluation result in the face portion of the person. As can be seen from the figure, the density diffusion method has less deterioration in image quality even in an image in which erroneous determination of image characteristics has occurred, and has obtained an evaluation result superior to that of the window method in both a character portion and a halftone portion. As described above, according to the present invention, it is possible to suppress the deterioration of the image quality at the pixel position where the image characteristic is erroneously determined, and the effect is clear from the subjective evaluation result shown in FIG.

It should be noted that the multi-value estimation method described in claim 2 of the present invention is a view of the method from another point of view, and is basically the same as that described in claim 1.

[0042]

As described above, according to the present invention,
When estimating the multi-valued image of the target pixel based on the pixel value and the weighting factor of each pixel in the reference region set around the target pixel to be multi-valued estimated, Since the weighting coefficient of each pixel in the reference area is distributed according to the image characteristic of each pixel, a window in which the weighting coefficient of each pixel in the reference area is fixed is selected from the image characteristic of only the pixel of interest. Of multi-valued images, which is a problem in the conventional method, can prevent the occurrence of singular points, which are the cause of density differences at pixel positions where image characteristics are erroneously determined, and improve the quality of estimated multi-valued images. Can contribute to

[Brief description of the drawings]

FIG. 1 is a conceptual diagram of a multi-level estimation method according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram for explaining density diffusion processing of a multi-value estimation method according to the present invention.

FIG. 3 is a conceptual diagram for explaining a difference between a multi-level estimation process according to a conventional window system and a multi-level estimation process according to a density diffusion system of the present invention.

FIG. 4 is a conceptual diagram for explaining a difference between a conventional window method and a multi-value estimation process using a density diffusion method according to the present invention when erroneous determination of image characteristics occurs.

FIG. 5 is a diagram showing a subjective evaluation result regarding a recording result of a multi-valued estimated image based on the conventional window method and the density diffusion method of the present invention.

FIG. 6 is a diagram showing each aspect of a reference window used in a conventional window system.

[Explanation of symbols]

 101 binary image 102 image characteristic discriminating unit 103 density diffusion unit 104 estimated multi-valued image

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int. Cl. ⁷ , DB name) H04N 1/40-1/409 H04N 1/46 H04N 1/60

Claims (3)

(57) [Claims] 1. A multi-value estimation method for estimating and outputting a multi-value image from a binary image having binary information of pixels, each of which is a component of the image, is either white or black. Image characteristic determining means for referring to the array of each pixel in the peripheral region for each pixel and determining the image characteristics of each pixel based on the array; and a reference region around the pixel of interest to be multi-valued estimated. A multi-valued image estimating means for setting and estimating a multi-valued image of the pixel of interest based on a pixel value and a weighting factor of each pixel in the reference area; and, when estimating the multi-valued image, the reference including the pixel of interest. A multi-value estimation method comprising: a density diffusion unit that distributes a weighting coefficient for each pixel in the reference area according to the image characteristics of each pixel in the area. 2. The multi-value estimation method according to claim 1, further comprising means for designating the reference region for a target pixel to be multi-valued. 3. The multi-value estimation method according to claim 1, wherein said image characteristic determining means determines said image characteristics at least for a character portion and a halftone portion.