Apparatus and method for processing image data with mixed background image and target image

The present invention relates to an apparatus and a method for processing image data in which a background image and a target image are mixed.

As an image processing device of this type, a check processing device described in Patent Document 1 is known. This check processing device optically scans securities such as checks and cash vouchers and acquires image data of the securities. In order to reduce the amount of image data, the original image data of the securities obtained by scanning (for example, grayscale image data) is subjected to binarization processing to create binary (black and white) image data of the securities. .. In this binary image data, it is necessary that various items to be described on the security (for example, bank name, account number, amount, payer name, signature, etc.) are clearly displayed. Here, securities such as checks usually have background images such as various patterns and patterns on the surface thereof, and the items described on the securities are often overlaid on the background image. .. The improvement disclosed in Patent Document 1 erases the background image from the original image data of the security and sets the binarization threshold to the image feature of the security so as to extract the image of the above-mentioned item to be read. The point is that it changes dynamically according to the situation. Specifically, before scanning the entire area of a security, a partial area of the security is scanned, and the image data obtained by the partial scan is used to obtain a histogram of the density (brightness) distribution of that partial area. , The binarization threshold is calculated based on the histogram.

<patcit num="1"><text>Japanese Unexamined Patent Publication No. 2004-123117</text></patcit>

<p> The technique of dynamically changing the binarization threshold based on the image features of the original image data disclosed in Patent Document 1 is for extracting a target image such as a description item on a securities from the original image data of a securities. It is very effective for. However, it is still difficult to solve some of the following problems with this technology alone.</p><p> First, securities statements often include image elements that are finer than the scan resolution (for example, extra-fine lines that are narrower than the one-pixel size of the scanned image data), and such extra-fine or fine. Image elements may be lost due to binarization processing. That is, since such ultrafine or fine image elements are represented by gray pixels whose density is not so high on the scanned image data, they are converted into white pixels (that is, erased) by the binarization process. )Sometimes.</p><p> Secondly, there are a wide variety of background images on securities, and the density distribution differs for each background image, and the density range may overlap with the density range of the target image (items described in the securities). In such a case, the background image and the target image cannot be distinguished by the density. As a result, a part of the background image is extracted as a black region together with the target image by the binarization process. When the target image overlaps the background image extracted as the black region, it is impossible to identify the target image on the binary image data.</p><p> Such a problem occurs not only in the use of securities reading but also in the use of reading other documents and other image usage when trying to distinguish the background image and the target image by binarization processing. Will do.</p><p> Therefore, an object of the present invention is to improve the accuracy of image processing for erasing a background image and extracting a target image from image data in which a background image and a target image coexist.</p><p> Another object of the present invention is to remove a background image from multi-gradation image data of securities and other documents to improve the accuracy of image processing for extracting an image of items described in the document.</p>

<p> The image processing apparatus according to the first aspect of the present invention receives multi-gradation original image data in which a background image and a target image are mixed, and the density distribution range of the background image and the density distribution range of the target image are one to the other. Based on the density distribution improving means for improving the density distribution of the original image data and generating the improved image data of multiple gradations, and receiving the improved image data and receiving the improved image data, based on the density distribution of the improved image data. , The binarization threshold calculation means for calculating the binarization threshold, the improved image data and the binarization threshold are received, and the improved image data is binarized using the binarization threshold to obtain a binar image. It is equipped with a binarizing means for generating data.</p><p> The image processing method according to the second aspect of the present invention receives multi-gradation original image data in which a background image and a target image are mixed, and the density distribution range of the background image and the density distribution range of the target image are one to the other. Based on the density distribution improvement step of improving the density distribution of the original image data to generate multi-gradation improved image data, and receiving the improved image data and receiving the improved image data, based on the density distribution of the improved image data. , The binarization threshold calculation step for calculating the binarization threshold, the improved image data and the binarization threshold are received, and the improved image data is binarized using the binarization threshold to obtain a binar image. It has a binarization step to generate the data.</p><p> According to this image processing device or method, the amount of overlap between the density distribution range of the background image and the density distribution range of the target image is reduced by performing the density distribution improvement processing on the multi-gradation original image data. The two images will be more clearly separated in the density range. The binarization threshold calculation and the binarization processing using the binarization threshold are performed using the improved image data for which the density distribution improvement processing has been completed. As a result, it is possible to obtain binary image data in which the background image is better removed and the target image is better extracted.</p><p> Therefore, it is important that the concentration distribution improvement process is performed before the binarization threshold calculation and the binarization process. In a preferred embodiment, the image sharpening or contrast enhancement process is adopted as the density distribution improving process, but this is an example, and other processing methods may be adopted.</p><p> The binary image data obtained by the binarization process may be further subjected to a noise removal process for removing so-called salt and pepper-like noise. In particular, when image sharpening or contrast enhancement processing is adopted as the degree distribution improvement processing, there is a high possibility that sesame salt-like noise will be generated. Therefore, further performing the sesame salt-like noise removal processing has good image quality. It is meaningful to obtain binary image data.</p><p> As a calculation method of the binarization threshold value, a method of detecting a junction between the density distribution range of the background image and the density distribution range of the target image in the improved image data and setting the binarization threshold value at this junction is adopted. be able to. As a method of detecting the joint portion, a histogram of the improved image data can be calculated, and the joint portion can be found from the slope of the histogram. For example, since the valley between the peak of the histogram of the background image and the peak of the histogram of the target image is often the junction between the two, a method of finding such a valley from the slope of the histogram can be adopted.</p><p> In a preferred embodiment, as a method of calculating the binarization threshold value, a histogram of the improved image data is calculated, inflection points satisfying a predetermined condition are searched from the histogram, and inflection points are binarized based on the searched inflection points. A method of determining the threshold is adopted.</p><p> The image processing apparatus according to the third aspect of the present invention receives multi-gradation original image data of a document in which the items to be described are printed or hand-written on a paper having a background image, and has a density distribution range of the background image. The density distribution improving means for improving the density distribution of the original image data to generate multi-gradation improved image data and the improved image data are received so that the density distribution range of the above items is far from one to the other. , The binarization threshold calculation means for calculating the binarization threshold based on the density distribution of the improved image data, the improvement image data and the binarization threshold, and the improvement using the binarization threshold. It is provided with a binarization means for binarizing image data and generating binary image data of the document.</p><p> The image processing method according to the fourth aspect of the present invention receives multi-gradation original image data of a document in which the items to be described are printed or hand-written on a paper having a background image, and has a density distribution range of the background image. In response to the density distribution improvement step of improving the density distribution of the original image data to generate multi-gradation improved image data and the improved image data so that the density distribution range of the above items is far from one to the other. , The binarization threshold calculation step for calculating the binarization threshold based on the density distribution of the improved image data, the improvement image data and the binarization threshold, and the improvement using the binarization threshold. It includes a binarization step of binarizing the image data and generating the binar image data of the document.</p><p> According to this image processing apparatus or method, it is possible to obtain binary image data from which the background image is better erased and only the description items are better extracted from the image data of the document.</p>

<p> According to the apparatus and method according to the first and second aspects of the present invention, the accuracy of image processing for erasing the background image and extracting the target image from the image data in which the background image and the target image coexist is improved. be able to.</p><p> According to the apparatus and method according to the third and fourth aspects of the present invention, an image for removing a background image from multi-gradation image data of a securities or other document to extract an image of a document description. The purpose is to improve the accuracy of processing.</p>

FIG. 1 is a block diagram showing a functional configuration of a main part of an embodiment of an image processing apparatus according to the present invention. This embodiment is an application of the present invention to a check reader for scanning checks and processing scanned image data of checks, but this is an example for illustration purposes and various other uses. Needless to say, the present invention can be applied to, for example, reading securities other than checks and various other documents, and various image processing application programs.

As shown in FIG. 1, the check reading device 20 includes an image reading unit 22, an image processing unit 24, and a reading control unit 26. The image reading unit 22 is an image scanner for a check having a known configuration, and optically scans the surface of the check set therein to display a multi-gradation image (for example, a grayscale image) of the check surface. The raster image data (hereinafter referred to as the original image data) 40 is output. The image processing unit 24 receives the original image data 40 output from the image reading unit 22, processes the received original image data 40 by a method according to the principle of the present invention, and displays the binary image data 46 (or 48) of the check. Is created, and its binary image data 46 (or 48) is output. The reading control unit 26 gives control signals 28 and 30 to the image reading unit 22 and the image processing unit 24, respectively, to control the operations of the image reading unit 22 and the image processing unit 24, respectively.

Hereinafter, the image processing unit 24 will be described in more detail.

The purpose of using this check reader 20 is to display characters representing various items to be described on the check (bank name, account number, payment amount, payer name, payee name, signature, various numbers, etc., and the entry position of those characters. The purpose is to generate binary image data for checks that clearly show the guide underline, frame line, etc.). Usually, some pattern or pattern is pre-printed on the surface of the check paper as a background, and the various items described above are printed or handwritten on the check paper. Therefore, in the original image data 40 of the check output from the image reading unit 22, an image of a background pattern or pattern (hereinafter referred to as a background image) and an image of the items to be described (hereinafter referred to as a target image) are mixed. In many cases, the latter overlaps the former, and the background image is useless for the purpose of using the check reading described above, and only the target image (image of various items described on the check) is desired. Therefore, the image processing unit 24 is configured to process the original image data 40 of the check according to the principle of the present invention, erase the background image from the original image data 40, and selectively extract only the target image as much as possible. ing.

In order to perform such image processing, the image processing unit 24 includes a density distribution improvement unit 32, a binarization unit 34, and a threshold value calculation unit 36. The original image data 40 from the image reading unit 22 is first input to the density distribution improving unit 32. The density distribution improvement unit 32 receives the original image data (multi-gradation image data) 40, and the original image data so that the density distribution ranges of the background image and the target image in the original image data 40 are separated from one to the other. Improve the concentration distribution of 40. Normally, in the original image data 40, the density of the target image is biased to a high density range, and the density of the background image is biased to a lower density range than the target image. Therefore, the bias is more remarkable in the density distribution improvement process. That is, the density distribution is improved or corrected so that the density of the target image is closer to the higher density direction and the density of the background image is closer to the lower density direction. As a result, the overlap of the density distribution ranges of the background image and the target image is reduced, and the background image and the target image are separated into different density ranges more clearly. Several methods can be adopted as a specific method for improving the density distribution, but in the present embodiment, a method of image sharpening or contrast enhancement is adopted as an example. The multi-gradation image data (hereinafter referred to as improved image data) 42 whose density distribution is improved, which is output from the density distribution improving unit 32, is then input to the threshold value calculation unit 36 and the binarization unit 34.

The threshold value calculation unit 36 dynamically determines the binarization threshold value 44 for distinguishing the background image from the target image dynamically based on the image features of the improved image data 42, particularly the features of the density distribution. That is, the threshold value calculation unit 36 creates a histogram of the density distribution of the entire region of the improved image data 42, and calculates a binarization threshold value 44 for distinguishing the background image and the target image based on the histogram. Since the density distribution improving unit 32 has already improved the density distribution of the improved image data 42 so that the background image and the target image are separated into different density ranges as much as possible, the threshold value calculation unit 36 has already improved the background image and the target image. The binarization threshold 44 can be determined so that the images can be better distinguished. The binarization threshold value 44 is input to the binarization unit 34.

The binarization unit 34 binarizes the improved image data 42 from the density distribution improvement unit 32 using the binarization threshold value 44 from the threshold value calculation unit 36. That is, in the improved image data 42, the region having a density equal to or higher than the binarization threshold value 44 is converted to a black region, and the region having a density lower than the binarization threshold value 44 is converted to a white region. As described above, in the improved image data 42 that is the target of the binarization process, the density distribution is improved so that the background image and the target image are separated into different density ranges as much as possible, and the binarization threshold 44 is set. , The background image and the target image of the improved image data 42 are defined to be well distinguished. Therefore, by the binarization process, most of the background image is converted into a white area and erased, and most of the target image is converted into a black area. Therefore, the binary image data 46 in which the target image is satisfactorily extracted can be obtained.

The binary image data 46 output from the binarization unit 34 is output from the image processing unit to the outside of the check processing device 20. As a modification, as shown by the dotted line in FIG. 1, a noise removing unit 38 is added downstream of the binarizing unit 34, and the binary image data 46 output from the binarizing unit 34 is used as the noise removing unit 38. In the noise removal unit 38, the noise contained in the binary image data 46 (for example, so-called sesame salt-like noise, that is, fine black dots scattered in the white region and fine black dots scattered in the black region) (White dots, etc.) may be removed, and the noise-removed binary image data 48 may be output to the outside of the check processing device 20.

Hereinafter, how the check image is changed by the image processing by the image processing unit 24 described above will be specifically described. For reference, FIG. 2 shows the change in the image when the original image data 40 is directly binarized without improving the density distribution. FIG. 3 shows changes in the image when the density distribution of the original image data 40 is improved and then the binarization process is performed in the above-described embodiment of the present invention.

FIG. 2A and FIG. 3A illustrate the original image data 40 of the same check. As described above, the original image data 40 is multi-gradation image data (for example, 256-gradation grayscale image data). In the illustrated example, the original image data 40 has a plurality of (for example, three) background images 50, 52, and 54, and their density ranges are different from each other. For example, the density range of the first background image 50 is low, the density range of the second background image 52 is intermediate, and the density range of the third background image 54 is high. The density ranges of these different background images 50, 52, 54 may partially overlap. Further, the original image data 40 includes a large number of target images (images of items described in the check) 60. The density range of the target image 60 is higher than the density range of the background image 54, but may partially overlap the density range of the background image 54. Many target images 60 are located on top of background images 50, 52, 54.

First, as a reference example, a case where the original image data 40 as shown in FIG. 2A is directly binarized without being subjected to the density distribution improvement processing will be described. In this case, the binary image data 70 as shown in FIG. 2B may be obtained. In this binary image data 70, of the three background images 50, 52, and 54, the first background image 50 having the lowest density is all converted to a white area and erased, but the third background image with a higher density has a higher density. In the second background images 54 and 52, a considerable part of the background images 54 and 52 is converted into a black region in the same manner as the target image 60, making it difficult to identify the target image 60 that overlaps the target image 60. Further, the guide lines and borders of the target image 60, which are thinner than the scan resolution, are not completely converted into the black region and are partially lost.

For example, the following circumstances can be considered as the cause of such a problem. For example, if the density range of the target image 60 and the density range of the third background image 54 partially overlap, it is impossible to distinguish the target image 60 from the background images 54 and 52 in the density range. Further, the density range of the second background image 52 partially overlaps the density range of the third background image 54, and the target image 60 and the third and second background images 54 are displayed on the density distribution histogram. Suppose 52 is like a mountain. Then, the binarization threshold value may be set between the density range of the second background image 52 and the density range of the first background image 50. Further, if the guide lines and borders in the target image 60 are finer than the scan resolution, the density of the guide lines and borders on the original image data 40 may be lower than the binarization threshold. When these circumstances overlap, the above-mentioned problems occur.

On the other hand, as illustrated in FIGS. 3A, B and C, according to this embodiment, the above-mentioned problems are solved to a considerable extent.

That is, the original image data 40 as shown in FIG. 3A is further subjected to the density distribution improvement processing. In this embodiment, image sharpening or contrast enhancement processing is performed as the density distribution improvement processing. When this processing is performed, the pixels having a higher density than the other surrounding pixels on the original image data 40 are corrected so that the density is higher. On the other hand, a pixel having a density lower than that of other surrounding pixels is modified so that the density is even lower. In the target image 60 (especially characters and line segments such as those described in checks), many of the pixels that make up the target image 60 have a higher density than the surrounding area, so that the target image 60 is mostly one layer. It is corrected to a high concentration and emphasized. On the other hand, in the background images 50, 52, and 54, pixels having a higher density than the surroundings and pixels having a lower density are mixed and dispersed, and even if the area appears to have a higher density to the naked eye, the pixels having a lower density than the surroundings are there. Are distributed in large numbers. Therefore, in the background images 50, 52, and 54, the region where the density was originally low becomes even lower. Also, in the background images 50, 52, and 54, where the density was originally high, the low density part increased and the points that were higher than the surroundings were emphasized, just like the pointillism or sesame salt image. High-density dots are scattered on the low-density ground. However, this pointillistic or sesame salt-like background image portion can be easily distinguished with the naked eye from the target image 60 in which the whole is deeply emphasized. Looking at this improvement from the viewpoint of density distribution, the density distribution of the target image 60 is corrected to shift to higher density, and the density distribution range of background images 50, 52, 54 is corrected to shift to lower density. Therefore, the overlap of the two concentration ranges is reduced, and the two concentration ranges are separated more clearly. Looking at this in the form of a histogram of the density distribution (see Histogram 110 illustrated in Fig. 9 below), the peaks of the target image 60 exist in the high density range, and the peaks of the background images 50, 52, and 54 are present. It exists in a low density range, and there is a clearer valley between the two mountains than in the original image data 40, and both mountains are original. It will be more clearly separated than when the image data is 40. As a result, the improved image data 42 as shown in FIG. 3B is obtained. In this improved image data 42, the densities of the background images 50, 52, and 54 are lower, the densities of the target image 60 are higher, and the thin guide lines and borders are also dark and clear.

Next, the binarization threshold is determined based on the improved image data 42. In that case, on the histogram of the density distribution, the peaks of the target image 60 and the peaks of the background images 50, 52, and 54 exist separately with a valley in between. The valley of this histogram corresponds to the junction between the density distribution range of the target image 60 and the density distribution range of the background images 50, 52, and 54, and a binarization threshold value can be set in this valley (joint portion). ..

Next, the improved image data 42 is subjected to binarization processing using the binarization threshold value set in this way. In the improved image data 42, many of the pixels constituting the background images 50, 52, and 54 have a density lower than the binarization threshold value, so that they are converted into white pixels, while many of the pixels constituting the target image 60 are converted to white pixels. , Since it has a density higher than the binarization threshold, it is converted into a black pixel. As a result, binary image data 46 as shown in FIG. 3C is obtained, in which most of the background images 50, 52, 54 are erased and most of the target image 60 is extracted.

Even with this processing, it is actually difficult to completely erase the background images 50, 52, and 54 and completely extract the target image 60. So-called salt and pepper-like noise 62 may remain on the binary image data 46. As described above, one of the main causes of such noise 62 is that the areas where the background images 50, 52, and 54 were originally high in density are pointillized or drawn by the density distribution improvement process (particularly, the image sharpening process). It was like a satin hatching. That is, high-density dots are scattered in the area that is in a state such as pointillism or satin hatching, and are converted into black dots, that is, salt-and-pepper-like noise 62 by the binarization process. Such noise 62 does not cause a substantial problem when identifying the target image 60 on the binary image data 46, and may be left as it is. However, the above-mentioned noise removing treatment may be further performed to remove the salt and pepper-like noise 62.

Hereinafter, the specific methods of the binarization threshold calculation process and the noise removal process described above will be described.

FIGS. 4, 5 and 6 show some examples of methods for improving the density distribution, in particular, the image sharpening process adopted in this embodiment. Figure 4 shows an example of a basic image sharpening method.

As shown in FIG. 4, one pixel of interest 80 is selected from the original image data 40, and a processing area 82 having a predetermined matrix size composed of the pixel of interest 80 and adjacent pixels around it is selected. In the example of FIG. 4, as the processing area 82, an area of a 3 pixel × 3 pixel matrix centered on the pixel of interest 80 is selected. It is assumed that each of the pixels in the processing area 82 has a pixel value (the lower the pixel value, the higher the density) a to i as shown in the figure. An image sharpening filter 84 having the same matrix size as the processing area 82 is prepared in advance. The image sharpening filter 84 has a coefficient corresponding to each pixel in the processing area 82, and the coefficient for the pixel 80 of interest is m, and the coefficient for the four adjacent pixels of the pixel 80 of interest is -k. , The coefficient for four adjacent pixels of the pixel 80 of interest located in the diagonal direction is, for example, zero. Here, the coefficient m and the coefficient k are positive numbers such that m-4 × k = 1. This image sharpening filter 84 is applied to the processing area 82. As a result, the image sharpening calculation of s = e × m-(b + d + f + h) × k is performed, and the original density value e of the pixel of interest is converted into the improved density value s.

According to this calculation, a pixel having a higher density than the surrounding adjacent pixels is corrected to have a higher density, and a pixel having a lower density than the surrounding adjacent pixels is corrected to have a lower density. .. That is, the contrast between the adjacent pixels is emphasized. All the pixels in the original image data 40 are sequentially selected as the pixel of interest 80, and the above image sharpening calculation is performed each time, whereby the original image data 40 is converted into the improved image data 42.

FIG. 5 shows an example of tone curve processing that can be supplementarily adopted in image sharpening processing. In FIG. 5, the pixel value indicates the brightness value, so that the pixel value 0 indicates black, the pixel value 255 indicates white, and the lower the pixel value, the higher the density.

In the tone curve processing, a tone curve 86 having the characteristics shown in FIG. 5 is prepared in advance. The input pixel of the tone curve 86 is a pixel of the original image data 40. This tone curve 86 is applied to all the pixels of the original image data 40. As a result, in the original image data 40, all the regions in the predetermined extremely high density range (the range in which the pixel value is n1 or less) near black are converted to the highest density black (pixel value 0), and the predetermined poles near white are converted. All the regions in the low density range (the range where the pixel value is n2 or more) are converted to the lowest density white (pixel value 255). There is no change in the density in the other density range (the range where the pixel value is larger than n1 and smaller than n2). This tone curve processing can be performed before applying the image sharpening filter 84 shown in FIG. 4 (or the image sharpening filter 94 described below with reference to FIG. 6).

FIG. 6 shows another example of the image sharpening process.

As shown in FIG. 6, the pixel of interest 90 is selected from the original image data 40, and includes the pixel of interest 90 and a plurality of pixels surrounding the pixel of interest 90 at a position two or more pixels away from the pixel of interest 90. The processing area 92 is selected. This processing area 92 has a matrix size of at least 5 pixels × 5 pixels. An image sharpening filter 94 having the same matrix size as the processing area 92 is prepared in advance. The image sharpening filter 94 has a coefficient for the pixel of interest and a coefficient for each of a plurality of pixels surrounding the pixel of interest 90 at the outermost position of the processing region 92. The coefficient for the pixel of interest is m2, the coefficient for the four pixels located in the vertical and horizontal directions of the pixel of interest 90 is -k2, and the coefficient for the other pixels is zero. Here, the coefficient m2 and the coefficient k2 are positive numbers such that m2-4 × k2 = 1. Similar to the example shown in FIG. 4, all the pixels in the original image data 40 are sequentially selected as the pixel of interest 90, and the image sharpening filter 94 is applied to the processing area 92 centered on the pixel of interest 90 each time. Will be done.

Assuming that the pixels in the processing area 92 have pixel values (the lower the pixel value, the higher the density) a to i as shown in the figure, s = e × m2-(s = e × m2-( The image sharpening calculation of b + d + f + h) × k2 is performed, and the original density value e of the pixel of interest is converted to the improved density value s.

By applying the image sharpening filter 94 shown in FIG. 6, pixels having a higher density than the surrounding pixels separated by a certain distance are corrected so that the density is higher than that of the surrounding pixels separated by a certain distance. Pixels with a low density are modified to have a lower density. That is, the contrast between pixels separated by a certain distance is emphasized.

The image sharpening filter 94 shown in FIG. 6 may be used in combination with the image sharpening filter 84 shown in FIG. 4, or may be used in place of the image sharpening filter 84 shown in FIG. Further, a plurality of filters having different matrix sizes, which are the image sharpening filters 94 shown in FIG. 6, may be prepared and used in combination. Alternatively, by preparing one image sharpening filter 94 shown in FIG. 6 and setting a coefficient for the pixel at the intermediate position omitted by the symbol ... in FIG. 6, FIG. 4 The image sharpening filter 84 shown in 1 and other image sharpening filters of different matrix sizes may be incorporated into this one image sharpening filter 94.

7 and 8 are diagrams showing how the density distribution of the original image data 40 is changed by performing the density distribution improvement process such as the image sharpening process described above.

FIG. 7 shows an example of the density distribution histogram 100 of the target image and the density distribution histogram 102 of the background image in the original image data 40 before the density distribution improvement processing is performed. Both the histogram 100 and the histogram 102 partially overlap each other, and therefore, it is difficult to distinguish the target image and the background image well by the binarization process as it is.

FIG. 8 shows an example of the density distribution histogram 104 of the target image and the density distribution histogram 106 of the background image in the improved image data 42 after the density distribution improvement processing is performed on the original image data 40 having the histogram of FIG. Shown. Compared with the original image data 40, both histograms 104 and 106 are shifted in the higher density direction and the lower density direction, respectively, and the area of the overlapping portion between the two histograms 104 and 106 is reduced and separated well. Therefore, if the binarization threshold Pth is set near the end of the histogram 104 on the low density side of the target image, the target image and the background image can be skillfully distinguished by the binarization process.

FIG. 9 shows an example of a processing method for calculating the binarization threshold value based on the improved image data 42.

In the binarization threshold, first, a histogram of the density distribution of the improved image data 42 is calculated. As a result, a histogram 110 as illustrated in FIG. 9 is obtained. This histogram 110 has a shape obtained by adding the histogram 104 of the density distribution of the target image and the histogram 106 of the density distribution of the background image as illustrated in FIG. When calculating the histogram 110, in order to eliminate the influence of local frequency fluctuations, frequency averaging processing is performed in a small concentration range. That is, in the original histogram simply calculated from the improved image data 42, each pixel value is sequentially selected as the pixel of interest, and for each pixel of interest, a predetermined small density range centered on the pixel of interest (for example, the pixel of interest). The average value of the frequencies (within the range of pixel values ± 7 values) is calculated, and the average value is taken as the frequency at the pixel value of interest. As a method of calculating the average value, a simple average or a weighted average may be used. As a result, the fine power fluctuations existing in the original histogram are removed, and the histogram 110 as shown in the figure having a shape in which the power changes smoothly is obtained. By performing the above averaging calculation, the effective range of the histogram 110 is slightly narrower than the total density range from the pixel values 0 to 255. For example, when the density range of the averaging calculation is the range of the pixel value of interest ± 7 values, the effective range of the histogram 110 is from the pixel value 7 to the pixel value 248.

Next, the slope f'(i) of the histogram 110 (the ratio of the change in frequency to the change in pixel value) is calculated for each pixel value i. For example, the following formula is used as the formula for calculating the slope f'(i).

<maths num="1"><img file="JP2007028362A_D0001.tif" /></maths>Here, f (i) is the frequency at the pixel value n (the number of pixels having the pixel value n). According to this formula, f'(i) of each pixel value i has the minimum distance error from the five-point frequency plot in the range of pixel value i ± 2 centered on the pixel value i by the least squares method. It is given as the slope of a straight line to be converted. It should be noted that this is an example in which the range of the pixel value i ± 2 is broken, and a wider or narrower range may be used.

Next, the binarization threshold Pth is calculated by the following steps (1) to (4).

(1) Search for black inflection point Pmin As shown by arrow 112 in Fig. 9, the histogram 110 is displayed while incrementally incrementing the pixel value i from the minimum value (that is, from the maximum density value to the lower density). (While tracing), check the slope f'(i) of the histogram 110 at the pixel value i, and the following conditions, f'(i) × f'(i + 1) 0 and f'(i) &lt; Find the pixel value i that satisfies 0 and f'(i + 1) 0. If a pixel value i satisfying this condition is found, the found pixel value i is set as the black inflection point Pmin. That is, the black inflection point Pmin is an inflection point at which the slope tends to change from downward to upward or horizontal when the histogram 110 is traced from the maximum density value in the downward direction.

As can be seen by comparing FIGS. 9 and 8, the black inflection point Pmin set as described above is near the low density endpoint 105 of the histogram 104 of the target image shown in FIG. 8, among others. It is highly possible that it exists at a position slightly off the end point 105 to the higher concentration side.

If the black inflection point Pmin cannot be set by the above method (that is, if the pixel value i satisfying the above conditions is not found), the preset pixel value U is set to the black inflection point. Let it be Pmin. From experience, this pixel value U is a pixel value at which it is considered that the end point on the low density side of the histogram 104 of the target image exists in the vicinity, especially at a position slightly lower than that, and in the case of check reading, it is a pixel value. For example, the value can be set to a value near the pixel value 64 of the high density side quarter of the entire pixel range 0 to 255.

(2) Search for the white inflection point Pmax As shown by arrow 114 in Fig. 9, the histogram 110 is displayed while sequentially decrementing the pixel value i from the maximum value (that is, from the minimum density value to the lower density direction). (While tracing), check the slope f'(i) of the histogram 110 at the pixel value i, and the following conditions, f'(i-1) × f'(i) 0 and f'(i)&gt; Find the pixel value i that satisfies 0 and f'(i-1) 0. If a pixel value i satisfying this condition is found, the found pixel value i is set as a temporary white inflection point. That is, the temporary white inflection point is an inflection point at which the slope tends to change from downward to upward or horizontal when tracing the histogram 110 in the direction of increasing from the lowest density value.

Next, as shown by arrow 116 in FIG. 9, the trace direction is reversed from the temporary white inflection point, the pixel value i is incremented, and the slope f'(i) is checked. Find the pixel value i that first satisfies the condition f'(i)&gt; Y. Here, the value Y is a predetermined positive value indicating a moderately gentle positive slope f'(i). If a pixel value i satisfying this condition is found, the found pixel value i is set as the white inflection point Pmax.

When the white inflection point Pmax is set by the above method, the following additional conditions are also taken into consideration. That is, the additional condition is that the cumulative value of the frequency of the histogram 110 in the pixel value range on the lower density side (the side where the pixel value is larger) is smaller than the preset numerical value V. Is a condition that the white inflection point Pmax is not set. Here, the numerical value V is a value that is empirically considered to be appropriate as the total number of pixels of the background image or a value slightly less than that, and in the case of a check, for example, about 75% of the total number of pixels of the entire check image. Can be a value. By adding this additional condition, the white inflection point Pmax can be set at a position farther from the low density end point 105 of the histogram 104 of the target image shown in FIG. 8 to the low density side. can avoid.

As can be seen by comparing FIGS. 9 and 8, the white inflection point Pmax set as described above is near the low density endpoint 105 of the histogram 104 of the target image shown in FIG. 8, among others. It is highly possible that it exists at a position slightly off the end point 105 to the lower concentration side. On the other hand, as described above, the black inflection point Pmin is likely to exist at a position slightly deviated to the high density side from the end point 105 on the low density side of the histogram 104 of the target image shown in FIG. .. Therefore, it is highly possible that there is an endpoint 104 on the low density side of the histogram 104 of the target image between the black inflection point Pmin and the white inflection point Pmax, and there is a binary value in the vicinity of the endpoint 104. If the inflection point Pth is set, the target image and the background image can be distinguished relatively well by the binarization process.

In another way, searching for the black inflection point Pmin and the white inflection point Pmax as described above is the boundary or junction between the histogram 104 of the target image and the histogram 106 of the background image. Is substantially equivalent to detecting. Then, if the binarization threshold value Pth is set at this joint portion, the target image and the background image can be relatively well distinguished by the binarization process.

(3) Calculation of binarization threshold Pth A. When the white inflection point Pmax can be set If Pmin Pmax, then Pth = Pmin + (Pmax-Pmin) × W. Here, the coefficient W is a predetermined positive number less than 1, and can be, for example, a value near 0.5. In short, the binarization threshold Pth is set between the black inflection point Pmin and the white inflection point Pmax.

On the other hand, if Pmin&gt; Pmax, then Pth = Pmax × X. Here, the coefficient X is a predetermined positive number of 1 or more, and can be, for example, a numerical value of 1 or more and less than 2. In short, the binarization threshold Pth is a position slightly lower than the white inflection point Pmax (in many cases, it is a position between the black inflection point Pmin and the white inflection point Pmax). Is set to.

(4) Correction of binarization threshold Pth The cumulative value of the frequency of the histogram 110 in the pixel value range on the higher density side (smaller pixel value) than the binarization threshold Pth is from the preset numerical value Y. When it becomes large, the binarization threshold Pth is moved to a point on the higher concentration side, and the binarization threshold Pth is corrected so that the cumulative value becomes equal to or less than the numerical value Y. However, the binarization threshold Pth is not set to a pixel value smaller than the predetermined lower limit pixel value Z (a pixel value having a density higher than the lower limit pixel value Z is not set). Here, the numerical value Y is a value that is empirically considered to be appropriate as the total number of pixels of the target image or a value slightly larger than that, and in the case of a check, for example, about 25% of the total number of pixels of the entire check image. Can be a value. Further, the lower limit pixel value Y is a pixel value that is empirically considered that the end point 105 on the low density side of the histogram 104 of the target image cannot exist on the higher density side (the side with the smaller pixel value). , The value can be slightly larger than the pixel value U (in the case of a check, for example, a value near the pixel value 64) used when setting the black side turning point Pmin described above.

By the above procedure, as can be seen by comparing FIGS. 8 and 9, in many cases, the binarization threshold Pth is set near the endpoint 105 on the low density side of the histogram 104 of the target image. Become. By binarizing the improved image data 42 using this binarization threshold Pth, in many cases, most of the background image is erased and most of the target image is extracted.

FIG. 10 shows an example of a noise removal processing method.

In this method, the entire area of the binarized image data is scanned, and the shading pattern 120 as shown in FIG. 10A and the shading pattern 124 as shown in FIG. 10B are searched. The shade pattern 120 and the shade pattern 124 are both typical shade patterns of so-called salt and pepper-like noise. That is, in the shading pattern 120, one pixel is black, and the pixels around the top, bottom, left, and right are all white, and represents black dot noise that exists isolated on a white ground. On the contrary, in the shading pattern 124, one pixel is white, and the pixels around the top, bottom, left, and right are all black, and represents white dot noise that exists isolated on the black ground. The found shading pattern 120 is converted into a white background pattern 122 consisting of all white pixels, whereby black dot noise is eliminated. Further, the found shade pattern 124 is converted into a black background pattern 126 composed of all black pixels, whereby white dot noise is eliminated.

Although the embodiments of the present invention have been described above, this embodiment is merely an example for explaining the present invention, and the scope of the present invention is not limited to this embodiment. The present invention can be carried out in various other aspects without departing from the gist thereof.

As a method of density improvement processing, a method other than the above-mentioned image sharpening or contrast enhancement can be adopted in combination with or in place of the image sharpening or contrast enhancement.

The histogram of the density distribution of the improved image data used in the binarization threshold calculation process may be created using all the pixels of the improved image data, but in order to reduce the processing load, a part of the improved image data may be created. It may be made using pixels. For example, a partial image area may be selected from the improved image data, and a histogram may be created based on the pixels of the partial image area. Alternatively, the pixel size of the improved image data may be reduced by a known method, and a histogram may be created based on the pixels of the reduced improved image data. Alternatively, representative pixels may be picked up from the improved image data at predetermined pixel intervals along the raster scan path, and a histogram may be created based on those representative pixels.

In the image of a check or the like securities processed in the above-described embodiment, the securities description item which is the target image is usually written in a high density color such as black or navy blue, while most of the background image is written. It is printed in a color with a lower density than the target image. However, the image processing of the present invention is effective even when the shading relationship between the target image and the background image is opposite to the above.

Further, when the density distribution range of the background image exists near the center of the entire density range of pixel values 0 to 255, and the density distribution range of the target image exists near both ends of the total density range, or vice versa. The present invention can also be applied to the case where there is a shade relationship of. In this case, instead of processing the entire density range of the pixel values 0 to 255 of the original image data together, some partial density ranges that make it easy to separate the density range of the background image and the target image are included. The process of the present invention can be applied to each of the set partial concentration ranges. As an example, the density distribution range of the background image is an intermediate density range with a pixel value of about 30 to 220, and the density distribution range of the target image is a high density range with a pixel value of 0 to 50 and a low density range with a pixel value of about 200 to 255. It is assumed that there is original image data that is within the range. In this case, for example, the first processing density range 0 to N1 (here, pixel value N1 &lt;200) and the second processing density range N2 to 255 (here, pixel value N2&gt; 50) are set, respectively, and the original is set. For each of the first processing density range 0 to N1 and the second processing density range N2 to 255 of the image data, the image processing of the present invention (that is, density distribution improvement, binarization threshold determination, and binarization) is performed separately. A series of processing) can be performed. At that time, when processing the first processing density range 0 to N1, all the pixels having a density lower than the pixel value N1 in the original image data are regarded as the pixel value N1, while the second processing density range N2 ~ When processing 255, all pixels having a density higher than the pixel value N2 in the original image data can be regarded as the pixel value N2. In the binary image data obtained as the processing result of the first processing density range 0 to N1, the portion of the target image belonging to the high density range 0 to 50 is separated from the background image and extracted. .. On the other hand, in the binary image data obtained as the processing result of the second processing density range N2 to 255, the portion of the target image belonging to the low density distribution range 200 to 255 is separated from the background image and extracted. It will be.

<figref num="1">The block diagram which shows the functional structure of one Embodiment of the image processing apparatus of this invention.</figref><figref num="2">As a reference, the figure which shows the change of the image when the original image data 40 is directly binarized without performing the density distribution improvement processing.</figref><figref num="3">The figure which shows the change of the image at the time of performing the density distribution improvement on the original image data 40, and then performing the binarization processing in the embodiment of this invention.</figref><figref num="4">The figure which shows the example of the basic method of the image sharpening process which is one of the density distribution improvement processes.</figref><figref num="5">The figure which shows the example of the tone curve conversion process which can be adopted auxiliary in the image sharpening process.</figref><figref num="6">The figure which shows another example of the image sharpening process.</figref><figref num="7">The figure which shows the histogram of the density distribution of the target image and the background image in the original image data.</figref><figref num="8">The figure which shows the histogram of the density distribution of the target image and the background image in the improved image data.</figref><figref num="9">The figure which shows the method example of the binarization threshold calculation processing.</figref><figref num="10">The figure which shows the method example of the noise removal processing.</figref>

Code description

20: Check reader, 22: Image reader, 24: Image processing unit, 26: Reading control unit, 32: Density distribution improvement unit, 34: Binarization unit, 36: Threshold calculation unit, 40: Original image Data, 42: Improved image data, 44: Binary threshold, 46: Binary image data, 48: Binary image data after noise removal, 50,52,54: Background image, 60: Target image, Pmin: Black Side turning point, Pmax: White side turning point, Pth: Binarization threshold.