JP5830338B2 - Form recognition method and form recognition apparatus - Google Patents

Description

  The present invention relates to an image processing technology such as an OCR (Optical Character Reader) and the like, and in particular, when a character color has a difference in density and the color to be removed is similar to the character color, the character Related to dropout technology that leaves only the color of the.

  A binarization process for generating a monochrome binary image from a color image is a basic process as a pre-process for a form OCR. The form has pre-printed ruled lines and characters called pre-printing, which becomes an obstacle when recognizing characters to be read. Therefore, the process of erasing, that is, whitening, the preprinted part in the binarization process of the form image is an essential function for improving the reading accuracy. Whitening of pre-printing using color information is called color dropout.

  As a typical prior art of color dropout processing, Patent Document 1 discloses that a color image is picked up, a color having the highest noise component luminance is selected from the three colors of RBG, and then a pixel in that color is selected. The method of binarizing using the gray value of the is described. Patent Document 2 describes a method of extracting only the character color after estimating the character color of the stamp so as to correspond to a different stamp color and pre-print color for each form. As another method, there is a method in which either red or blue is designated as a noise component, and an image captured with a light source having the same color as the noise component is used for character recognition. For example, if a form with black characters and red ruled lines is imaged with a red light source, the ruled line portions do not appear in the image. The same applies to the blue system.

JP 2003-196292 A JP 2010-244372 A

  An example of a form that requires color dropout processing is a receipt slip. Receipt slips have preprints such as ruled lines and characters, printed characters and handwritten characters are written, and date stamps are stamped as proof of deposit and withdrawal. Of these, reading date stamp characters has a problem that differs from normal written characters in color dropout processing. The ink color of the date stamp differs from red to blue to black for each form, and many crossings with pre-prints of various colors and shapes occur.

  Furthermore, due to the influence of how to apply and push the date stamp, the characters in the date stamp have a greater density difference than the normal characters written on the form. In addition, date stamps are often imprinted with inks of the same color as pre-printing. There is a trade-off between the difference in density of characters and pre-printing of similar colors. In general, in order to allow a difference in density of the character color, the dropout threshold is lowered in order to extract a thin portion of the character. This may cause noise components and the like to be extracted at the same time, making separation from pre-printing difficult. On the other hand, in order to drop out preprints of similar colors, the threshold value becomes strict, and characters with large shade differences cannot be extracted.

  The method described in Patent Document 1 is based on the premise that characters and pre-printing are different colors, and therefore cannot be applied when characters and pre-printing have similar colors. In addition, the method described in Patent Document 2 dynamically determines a dropout threshold, but cannot cope with a case where a character with a large shade difference and a ruled line of the same color coexist. In addition, when either red or blue is specified as the noise component and an image captured with a light source of the same color as the noise component is used for character recognition, when forms with different stamp colors and pre-print colors are mixed Is not applicable.

  In view of the problems in the conventional technology as described above, the object of the present invention is to correctly extract a character of a pixel with a small amount of ink in dropout for a case where color unevenness is large, and to perform pre-printing around the character. Even if there is a form with a large number of characters and a form with a large difference in density of characters, it is to drop out appropriately and recognize characters.

  Here, “color unevenness” refers to a state in which, for example, when a date stamp is imprinted on a form or the like, the color of characters or the like blurs or fades due to the amount of ink or the amount of power, and the difference in shade is large. .

  The image processing method according to the present invention is a dropout image processing method that leaves only character components from the processing target image, the step of inputting the processing target image, the step of color clustering the pixels of the processing target image, and the color clustering A background color based on the image, a step of selecting a stamp color from the pixels of the processing target image, a color space conversion of the pixels of the processing target image, and a shading for dealing with color unevenness from the pixels of the processing target image An image generating step; a step of correcting the gradient of the grayscale image; and a step of binarizing the grayscale image subjected to the gradient correction.

  In the present invention, in the dropout for a form in which the character component color and the noise component color are indefinite, even if the color unevenness is large, the character color can be correctly left. For this reason, the date character string can also be correctly extracted.

1 is a configuration diagram of an image processing apparatus according to an embodiment of the present invention. It is a figure which shows the outline of the dropout process of conventional embodiment. It is a figure which shows the outline of the dropout process of embodiment of this invention. It is a figure which shows an example of the process of a stamp date reading of embodiment of this invention. It is a flowchart which shows the dropout process which outputs the binary image for color nonuniformity of embodiment of this invention. It is a flowchart which shows the stamp color selection process of embodiment of this invention. It is a figure which shows the example which a date stamp drops out by the dropout process in the conventional stamp color selection. It is a flowchart which shows the circumference color selection process of embodiment of this invention. The flowchart which shows the stamp color determination process of embodiment of this invention. It is a flowchart which shows the grayscale image production | generation process for color unevenness of embodiment of this invention. It is a figure which shows the outline of the hue frequency distribution of a date stamp, and a date stamp hue tolerance | permissible_range. It is a flowchart which shows the determination process of the pixel classification for color nonuniformity in the pixel of interest when the stamp color of the embodiment of the present invention is an achromatic color. It is a flowchart which shows the determination process of the pixel classification for color nonuniformity in the pixel to which attention is paid when the stamp color is a chromatic color according to the embodiment of the present invention. It is a figure which shows the outline of the binarization process of embodiment of this invention. It is a flowchart which shows the 2nd process of stamp date reading of embodiment of this invention. It is a flowchart which shows the dropout process which outputs the binary image for pre-printing / background pattern removal of embodiment of this invention. It is a flowchart which shows the grayscale image production | generation process for pre-printing and a background pattern removal of embodiment of this invention. It is a flowchart which shows the determination process of the pixel classification for pre-printing / background pattern removal in the pixel of interest when the stamp color is an achromatic color according to the embodiment of the present invention. It is a flowchart which shows the determination process of the pixel classification for pre-printing / background pattern removal in the pixel of interest when the stamp color is a chromatic color according to the embodiment of the present invention. It is a figure which shows the outline of the binary image output result by the presence or absence of ruled line removal when the date stamp of embodiment of this invention is stamped on the ruled line of similar color. It is a flowchart which shows the 3rd process of stamp date reading of embodiment of this invention. It is a flowchart which shows the dropout process which outputs the two binary images of the binary image for color unevenness of embodiment of this invention, and the binary image for pre-printing / background pattern removal. It is a flowchart which shows the 4th process of stamp date reading of embodiment of this invention. It is a flowchart which shows the 5th process of stamp date reading of embodiment of this invention. It is a flowchart which shows the 6th process of stamp date reading of embodiment of this invention. It is a flowchart which shows the 7th process of stamp date reading of embodiment of this invention. It is a flowchart which shows the 8th process of stamp date reading of embodiment of this invention. It is a figure which shows the 1st example of the user interface which confirms and corrects the stamp date reading result of embodiment of this invention. It is a figure which shows the 2nd example of the user interface which confirms and corrects the stamp date reading result of embodiment of this invention. It is a figure which shows the 3rd example of the user interface which confirms and corrects the stamp date reading result of embodiment of this invention.

<First embodiment>
Embodiments of the present invention will be described with reference to the drawings. In addition, this invention is not limited by the following description.

  FIG. 1 is a diagram showing a configuration of a form recognition system according to an embodiment of the present invention. The form recognition system includes an input device 10, an image input device 20, an image processing device 30, a dictionary 40, a display device 50, and an image database (DB) 60. The input device 10 is a device such as a keyboard and a mouse for inputting commands and code data to the image processing device 30. The image input device 20 is a device such as a scanner for inputting a form as image data to the image processing device 30. The image processing apparatus 30 is a computer that detects a reading area of a form input by the image input apparatus 20 and performs a dropout process, and includes a CPU, a memory, and a storage device (not shown). The image processing apparatus can also execute processing such as character recognition of the dropout image. The dictionary 40 is a dictionary database that is referred to when the image processing apparatus 30 recognizes a form. Specifically, the dictionary 40 is referred to when a character recognition dictionary that is referred to when the image processing apparatus 30 recognizes characters, a knowledge dictionary that is referred to when character string matching is performed, or a form reading area. Stores form information and the like. The display device 50 is a device such as a display that displays the result of the form recognized by the image processing device 30. The image DB 60 stores image data input to the image processing device 30 by the image input device 20. The image DB 60 may store in advance image data to be recognized by the image processing apparatus 30 by the image input apparatus 20.

  The present invention may be implemented in a normal computer by software having the same function as the image processing apparatus 30.

  Before showing a specific example of the dropout method realized by the present invention, an outline of a conventional dropout processing result is shown in FIG. 2, and an outline of a processing result for color unevenness of the present invention is shown in FIG.

  FIG. 2 is a diagram showing an outline of a conventional dropout processing result as shown in Patent Document 2, for example. FIG. 2A shows an input image that is almost uniformly printed on the pre-print and is stamped uniformly. Reference numeral 201 denotes a region to be processed, 202 denotes a date stamp that is uniformly pressed without color unevenness, and 203 denotes pre-printing such as a ruled line on a form or a background pattern. A color dropout image (FIG. 2 (b)) in which only the color component 204 of the stamp remains is generated by dropping out pre-printing and ground pattern colors other than the stamp color from the input image of FIG. 2 (a). . When image correction such as tilt, binarization, and date row extraction are performed from the color dropout image of FIG. 2B, a date row binary image (FIG. 2C) having only the date row component 205 is generated. .

  FIG. 2D shows an input image with uneven color. The date stamp 207 in the processing target area 206 includes a thin pixel 208 and a dark pixel 209, and the date stamp as a whole has large color unevenness. When dropout is performed on the input image of FIG. 2D, it is determined that the thin pixel is not the stamp color, so that the color dropout image (FIG. 2) in which the thin pixel is dropped out and only the dark pixel 210 remains. 2 (e)) is generated. When image correction such as inclination is performed from the color dropout image of FIG. 2E, binarization, and date line extraction are performed, a date line binary image with only the date line component 211 missing a part of the date (FIG. 2 ( f)) is generated.

  As described above, the conventional dropout has a problem in that character components are lost when color unevenness is large. Here, the first problem of the present invention is to correctly extract a date character line without missing components due to dropout even when such color unevenness is large. In order to realize this, an outline of a solution to the problem in the present invention will be described. Details will be described with reference to FIG.

  (1) Stamp color selection: In the conventional method, the cluster having the next largest number of pixels after the background color is used as the stamp color. In many cases, pre-printing or copy-forgery-inhibited pattern color may be used as the stamp color. In the present invention, first, color clustering is performed using only the colors of the pixels near the circumference of the date stamp, and the cluster that is not close to the background color and has the highest frequency is selected as the circumferential color. Next, color clustering is performed in the region within the stamp, and among the resulting clusters, the color of the cluster having a certain frequency and closest to the circumferential color is set as the stamp color. By this method, it is possible to prevent erroneous selection of preprinted or tint block colors as stamp colors.

  (2) Color dropout: In the conventional method, the threshold value for pre-printing or dropping out the background pattern using not only the RGB information of the pixel but also the HSV color space is severe, so if the color unevenness is large, the ink color of the stamp is Thin pixels are dropped out and lost. In the present invention, in the HSY color space obtained by converting the RGB information and the hue frequency distribution in the processing target area, the hue, saturation, and luminance allowable range are dynamically obtained, and thus the color unevenness is large. However, it can leave the character color.

  (3) Binarization: In the conventional method, when the color unevenness is large, the pixel where the ink color of the stamp is light becomes a white pixel and is lost, and the character is blurred. In the present invention, when the binarization threshold is obtained for each partial area in the date line area and binarized, in a partial area where there are many pixels with a light ink color of the seal, thin pixels can be converted to black pixels, By generating a binary image of the logical sum of all binary images binarized for each partial area, the binary image in the date line area can suppress missing characters even if the color unevenness is large. .

  FIG. 3 is a diagram showing an outline of the processing result of dropout corresponding to uneven color according to the present invention. FIG. 3A shows an input image having the same color unevenness as FIG. In the color dropout corresponding to the color unevenness, the color dropout image (FIG. 3B) is generated without the thin pixel 302 being dropped out. When image correction such as inclination, binarization corresponding to color unevenness, and date line extraction are performed from the color dropout image of FIG. 3B, a date line component 302 date line binary image (FIG. c)) is generated.

  FIG. 3 (d) shows an input image which is uniformly stamped with almost no color unevenness on the same pre-print as in FIG. 2 (a). When color dropout corresponding to uneven color is performed on the input image of FIG. 3D, a color dropout image (FIG. 3E) in which pre-printed ruled lines and characters 303 remain is generated. When image correction such as inclination, binarization corresponding to color unevenness, and date line extraction are performed from the color dropout image of FIG. 3E, the date line of the date line component 304 in which preprinted ruled lines and character components 305 remain is obtained. A binary image (FIG. 3F) is generated.

  As described above, there is a problem that the ruled line and the background pattern component that have been removed by the dropout are left by dealing with the color unevenness. Here, the second problem in the present invention is that the date character line is correctly extracted in both cases even when the color unevenness is large or the case where there is a pre-print / background pattern of similar colors is mixed. is there. In order to realize this, the present invention adopts the following method. The first method is to select a dropout for color unevenness and a dropout for pre-printing / background pattern removal according to the situation and output a date character line binary image. The second method is to output two types of date character line binary images for dropout and for pre-printing / background pattern removal. Details will be described with reference to FIG.

  Hereinafter, an embodiment of a stamp date reading method and a stamp date reader to which the present invention is applied will be described. This is because even when the color imperfection of the date stamp is large, pixels of the light ink color of the stamp can be left.

  FIG. 4 is a diagram showing a processing flow of reading a stamp date to which the color dropout corresponding to color unevenness according to the present invention is applied. This is executed by the image processing apparatus (CPU) 30. Usually, it is realized as a program executed by the CPU, and such a program can be stored in various recording media, stored in a memory and executed by the CPU.

  The target color image is input in the image input 401, and the date stamp area is detected in the date stamp detection 402.

  Steps 404 to 406 are repeated for the number 403 of date stamps detected by the date stamp detection 402. In the color dropout 404 for outputting the color unevenness-corresponding binary image, the date line binary image is extracted from the processing target area including the target date stamp area, and is included in the dictionary 40 in the date character string recognition 405. Date character string recognition is performed using a character identification dictionary and a date notation dictionary. At 405, the date character string recognition result is set at 406, and the reading processing of the date stamp of interest is finished. When all date stamps detected in 403 have been read, the result is output in 407, and the stamp date reading process is terminated. A color dropout 404 for outputting a color unevenness-compatible binary image will be described with reference to FIG.

    FIG. 5 is a diagram showing a flow of color dropout processing for outputting a color unevenness-compatible binary image according to the present invention. In this color dropout process, a grayscale image in which only the stamp color is left is generated from the processing target area, and this is binarized to generate a binary image.

  The processing area selection process 501 is a process for selecting an area including a date stamp to be read from an image. This processing area is determined using the center coordinates and radius of the date stamp obtained by the date stamp detection 402.

  The color clustering process 502 is a process for color clustering each pixel in the processing area. In color clustering, each pixel in an image is mapped on a color space, and then colors close to each other are set to the same cluster based on a predetermined criterion. As a specific example of color clustering, the method disclosed in Patent Document 2 can be used. In order to facilitate subsequent processing, the result of color clustering may be sorted in the order of the number of pixels in the cluster.

  Background color selection processing 503 is processing for selecting a background color from the result of color clustering. As the background color, the color having the largest number of pixels (frequency) among the color clustering results is set as the background color. As an example of a selection criterion for a cluster including a background color, a cluster having the largest number of pixels is employed. This is because the area of the background is generally the largest among the date stamp, ruled line, noise, and background in the processing target area. Another example of a selection criterion for clusters including a background color is to select a cluster having the highest luminance. This is because the paper color has a higher brightness than the date stamp color. As the brightness of the cluster, the average brightness of the pixels included in the cluster or the brightness of the peak or center of the distribution can be used. Luminance Y is calculated as Y = 0.29891R + 0.58661G + 0.11448B with different weights assigned to RGB colors. On the other hand, when the frequency of the cluster including the date stamp color is higher than the cluster including the background color in the date stamp, the date stamp color may be erroneously determined as the background color in the above method. In order to prevent this, the background color may be selected according to the above reference example from the result of color clustering only the pixels outside the date stamp area.

  The date stamp color selection process 504 is a process for selecting a date stamp color from the result of color clustering. The stamp color selection process will be described with reference to FIGS.

  The color space conversion process 505 is a process for converting the pixel to be processed from the RGB color space obtained from the input device to the HSY space for use in the color unevenness gray image generation process 506. The HSY space expresses a color with hue (H) that is an index of color type, saturation (S) that is an index of vividness, and luminance (Y) that is an index of brightness. H and S are calculated by the same calculation formula as the widely used HSV color space. The reason why the HSY color space is used in the present invention is that the expression of colors using hue, saturation, and brightness is generally similar to the human color perception method compared to the expression using RGB. The reason for using the luminance (Y) instead of the lightness (V) as the brightness index is that the luminance is closer to a human brightness perception method. The color space conversion may use not only HSV and HSY, but also HSV color space, HSL color space, and the like. In addition, a CMY color space or a CMYK color space, which is a subtractive color expression used in the printing process, may be used.

  In this way, by using dimensions other than RGB, pixel classification processing, which will be described later, and the like become highly accurate. In the above example, the color space conversion is performed collectively and then the following processing is performed. However, the color space conversion is performed by calculating H, S, etc. from RGB each time a pixel is used. Processing may be performed.

  The color unevenness gray image generation processing 506 generates a gray image by leaving only the color suitable for color unevenness from the processing target area for the stamp color selected by the stamp color selection 504. The color unevenness gray image generation process will be described with reference to FIGS.

  The tilt correction 507 detects the line at the upper limit of the date area for the grayscale image generated by the grayscale image generation processing 506, obtains the tilt, and corrects this tilt to be upright. The merit of performing the inclination correction here is that a quantization error is less likely to occur after the inclination correction in the grayscale image as compared to the binarized image. Specifically, when an image including an oblique line or curve is corrected, a jagged line is hardly generated on the line. As an example of the inclination correction method, a binary method or a bicubic method can be used. In order to correct the inclination, it is necessary to detect the inclination. This can be obtained by detecting lines above and below the date line, for example, but may be obtained by other methods. The tilt may be detected within the tilt correction 507 or by other processing.

  The binarization processing 508 binarizes only the date line area from the grayscale image that has been tilt-corrected by the tilt correction 507, and generates a binary image. The binarization process 507 will be described with reference to FIG.

  The stamp color selection processing 504 will be described with reference to FIGS. FIG. 6 is a processing flow for selecting a stamp color. In the selection of the stamp color, in order not to mistakenly set the color of the pre-print or the background pattern, the circle color is first selected by the circle color selection process 601, and then the stamp color is determined by the stamp color determination process 602. In the conventional method, the cluster with the next largest number of pixels after the background color is used as the stamp color, so if the number of pre-printed or tint block color pixels is greater than the number of date stamp pixels for the processing target area, pre-printing is performed. And the background color may be used as the seal color. The reason for this will be described later with reference to FIG. In the present invention, first, color clustering is performed using only the colors of the pixels near the circumference of the date stamp, and the cluster that is not close to the background color and has the highest frequency is selected as the circumferential color. The reason why pixels near the circumference are used is that there are dense colors closest to the seal color around the circumference. Next, color clustering is performed in the region within the stamp, and among the resulting clusters, the color of the cluster having a certain frequency and closest to the circumferential color is set as the stamp color. By this method, it is possible to prevent erroneous selection of preprinted or tint block colors as stamp colors. The circumferential color selection processing 601 will be described with reference to FIG. 8, and the seal color determination processing 602 will be described with reference to FIG.

  FIG. 7 shows an example in which the stamp color is wrong with the color of pre-printing or background pattern. The input image (a) in FIG. 7 is an area selected by the target area selection processing 501 from the color image. For example, date stamp 701 is green with high luminance, pre-print 702 is brown with low luminance, and tint block 703 is brown with high luminance. In addition, in the result of color clustering, the frequency of pre-printing and ground pattern brown clusters may be higher than the date-marked green clusters. In this case, the stamp color is the color of the most frequently clustered color other than the background color in the result of color clustering, so the preprint 702 and the background pattern 703 are erroneously used as the stamp colors. In pixel determination, which is a subsequent process, only a color centered on the wrong stamp color is left, and a grayscale image is generated based on this color, and the grayscale image is as shown in FIG. 7B. Only 705 shade pixels remain. As a result of binarization and date line extraction of this grayscale image, a date character line binary image in which the pixels of the pre-printing and copy-forgery-inhibited pattern 706 are left as black pixels is generated (FIG. 7C). As described above, if the stamp color is selected by mistake, the date stamp is dropped out.

FIG. 8 is a diagram illustrating a processing flow of circumferential color selection. Step 801 designates the size of the radius r with a margin α. R _in is set to be α smaller than the radius r, and r _out is set to be larger than the radius r. This takes into account the radius r and center coordinates required for date stamp detection due to stamp distortion, quantization error, etc., and the width of the outer edge of the date stamp. Step 802 repeats the y coordinate of the target area, and step 803 scans the pixel of the target area by repeating the x coordinate of the target area. Step 804 adds the color of this pixel to the circumferential color list in Step 805 if the pixel is in a range in which the margin of the date stamp radius is included in the coordinates of the target area. When all the colors of the pixels in the range in which the margin of the date stamp radius is provided in the target area have been registered, color clustering is performed on the circumferential color list in step 806. In step 807, the representative color of the cluster with the highest frequency from the result of color clustering is set as the initial value of the circumferential color. In step 808, if the color distance between the circumferential color and the background color is less than the threshold value as the number of clusters as a result of the color clustering, a representative color of the cluster of interest is selected in step 809. In step 810, the circumferential color is Replace with the representative color of the cluster of interest.

  The reason why the circumferential color selected in this way is not used as the stamp color is that the date stamp may have uneven color depending on the method of stamping. This is because the color closest to the circumferential color in the entire date stamp is used as the stamp color.

FIG. 9 is a diagram showing a processing flow for determining the seal color. Step 901 is repeated for the y coordinate of the target area, and step 902 is performed for the x coordinate of the target area to scan the pixels of the target area. In step 903, if the coordinates x and y are within the radius of the date stamp, the color of the pixel at this coordinate is added to the stamp color list. At this time, if the color is close to the background color, a process of moving to the next coordinate without adding it to the stamp color list may be added. This is because the stamp color is different from the background color and does not need to be registered in the stamp color list.

  In step 905, color clustering is performed on the stamp color list obtained by scanning the entire target area. In step 906, the color of the cluster with the highest frequency is set as the initial value of the stamp color from the result of the color clustering. In step 907, steps 908 to 912 are repeated for the number of clusters in the clustering of the stamp color list. Step 908 selects the representative color of the cluster of interest. If the number of pixels of the cluster of interest is smaller than the threshold value in step 909, the loop exits in step 910. The reason why the number of pixels in the cluster is smaller than the threshold value is that it is different from the color of the date stamp. Step 910 exits the loop if the result of color clustering is sorted by the number of pixels in the cluster, but moves to the next cluster if not sorted. In step 911, if the distance between the representative color of the cluster of interest and the circumferential color obtained by the circumferential color selection 601 is less than the distance between the currently set stamp color and the circumferential color, the stamp color is Replace with the representative color of the cluster of interest. The stamp color is determined by the above processing.

  Next, the grayscale image generation processing 506 for dealing with color unevenness will be described with reference to FIGS.

  FIG. 10 is a diagram showing a processing flow for generating a gray image for dealing with uneven color. In the conventional method, the threshold for dropping out preprints and background patterns using not only the RGB information of the pixels but also the HSV color space is severe, so if the color unevenness is large, the pixels where the ink color of the stamp is light is dropped. Out and missing. In the present invention, in the HSY color space obtained by converting the RGB information and the hue frequency distribution in the processing target area, the hue, saturation, and luminance allowable range are dynamically obtained, and thus the color unevenness is large. However, it can leave the character color. In this process, the process is switched depending on whether the stamp color is an achromatic color or a chromatic color. An achromatic color is a color with a low saturation value and refers to a color such as white, gray, or black. A chromatic color refers to a color whose saturation is equal to or greater than a certain value. The reason for dividing achromatic colors into chromatic colors is that the criteria for color discrimination differ between achromatic colors and chromatic colors. The identification of achromatic colors is based on luminance and saturation. On the other hand, in chromatic colors, hue has the highest priority, followed by luminance and saturation.

  First, at 1001, the seal color is converted to HSY color space. In the subsequent processing, processing is performed using the RGB value of the stamp color and the HSY color space. In determination 1002, it is determined whether the stamp color is a chromatic color or an achromatic color.

  Whether the chromatic color or the achromatic color is determined is determined as an achromatic color when the saturation of the stamp color is less than the slide saturation threshold (1003). If the saturation of the stamp color is equal to or greater than the slide saturation threshold, a chromatic color is set (1005). Here, the slide saturation threshold is a threshold setting for changing the saturation threshold according to the luminance. In general, when the luminance is higher than when the luminance is low, the color change appears larger even if the saturation is the same. For this reason, the saturation threshold is set lower as the luminance increases.

  In the pixel classification (for achromatic color) 1004 for dealing with color unevenness for each pixel within the radius of the stamp, the pixels to be dropped out are classified for each pixel within the radius of the stamp. The pixel classification process is a process for classifying each pixel in the area into four types according to the distance from the stamp color and the background color for the purpose of dropout. The conventional pixel classification process is as described in Patent Document 2. The achromatic pixel classification 1004 in the present invention will be described with reference to FIG.

  A hue tolerance range calculation process 1006 for dealing with unevenness in color calculates a hue tolerance range that is a discrimination criterion for pixel classification. Since it is the discrimination of the chromatic stamp color, the hue is used as the discrimination criterion. The allowable range of hue must be changed depending on whether there is pre-printing in the stamped area or uneven color. The reason for this will be described with reference to FIG.

  FIG. 11A shows the frequency distribution of hues when there is pre-printing / background patterns. When the frequency distribution of the hue of the pixel is taken in the color image, the date spreads like a frequency distribution 1102 around the hue 1101 of the stamp color, and the pre-print / background pattern has a frequency distribution 1104 around the hue 1103. Suppose that In order to separate the pre-print / background pattern from the date stamp, it is necessary to divide the stamp color hue 1101 in the allowable range 1105. As an example of the calculation of the allowable range 1105, the frequency distribution 1102 may be treated as a normal distribution centered on the hue 1101, and the allowable range may be a hue that is within a deviation βσ (for example, β = 1). In order to reduce the calculation amount, the allowable range may be from the frequency of the hue 1101 until the frequency becomes 1 / m (for example, m = 2) or less.

  On the other hand, FIG. 11B shows a frequency distribution of hues when there is color unevenness. In the frequency distribution 1106 in the case where there is color unevenness, a part of the date mark may be hardened in the vicinity of the hue 1103 in some cases. In the range 1105, the component solidified in the vicinity of the hue 1103 is dropped out. Therefore, when there is uneven color, the allowable range is expanded to 1107. As an example of how to expand the allowable range, the frequency distribution 1106 may be treated as a normal distribution centered on the hue 1101 and the allowable range may be a hue that is within a deviation ασ (for example, α = 3). In order to reduce the calculation amount, the allowable range may be from the frequency of the hue 1101 until the frequency becomes 1 / n (for example, n = 10) or less. In the calculation of these allowable ranges, there may be a hue between the hue 1101 and the hue 1103 that has a frequency of 0 or a frequency that is less than 1 / n (for example, n = 10) from the frequency of the hue 1101. May be adapted by moving averages.

  In the pixel classification (for chromatic color) 1007 corresponding to color unevenness for each pixel within the radius of the stamp in FIG. 10, the pixel to be dropped out is classified for each pixel within the radius of the stamp. The pixel classification 1007 for chromatic colors will be described with reference to FIG. The gray value conversion processing 1008 generates a gray value image based on the luminance value of the pixel for each pixel classified by the pixel classification 1004 or 1007 by the method shown in the description of FIGS. In this gray value image, in order to make it easy to drop out colors other than the stamp color, the luminance values of the pixels determined to be the background and pre-printing are higher than the luminance area of the input image. For the determination that the color is other than the stamp color, the distance between the stamp color and the background color and the relationship with the adjacent pixels are used.

  12 and 13 are diagrams showing a pixel classification determination flow. FIG. 12 shows an achromatic color case and FIG. 13 shows a chromatic color case. In the pixel classification, the pixels in the target area are classified into four types of “background”, “character”, “removal candidate”, and “character candidate” based on the distance from the background color and the stamp color. Among these, the removal candidate assumes pre-printing or a background pattern, and clearly indicates a color different from the stamp. The character candidate is assumed to be pre-printed in the same color as the stamp, and is close to the stamp color, and therefore cannot be determined as a character or a removal candidate at this stage. This classification result is used to generate a gray image in the subsequent stage.

  FIG. 12 is a determination flow of pixel classification for dealing with color unevenness when the stamp color is an achromatic color. For each pixel within the stamp radius, pixel classification is performed according to this determination flow. If it is determined (1201) that the distance between the color of the pixel and the stamp color is equal to or greater than the threshold or the luminance of the pixel is higher than the luminance of the background color, the background flag is set (1202) for this pixel and the processing returns (1203). If it is determined that the distance between the pixel color and the stamp color is closer than the threshold (1204), the character flag is set (1205) for this pixel and the process returns (1206). When it is determined that the pixel saturation is equal to or less than the pixel slide saturation threshold and the luminance difference from the stamp color is equal to or less than the luminance difference threshold for uneven color (1207), a character flag is set for this pixel (1208). Then return (1209). In cases other than these determinations, a removal candidate flag is set (1210) for this pixel, and the process ends.

  FIG. 13 is a determination flow of pixel classification for color unevenness when the stamp color is a chromatic color. For each pixel within the stamp radius, pixel classification is performed according to this determination flow. If it is determined that the distance between the pixel color and the stamp color is equal to or greater than the threshold value or the luminance of the pixel is higher than the luminance of the background color (1301), the background flag is set for this pixel (1302) and the process returns (1303). If it is determined that the distance between the pixel color and the stamp color is closer than the threshold (1304), the character flag is set (1305) for this pixel and the process returns (1306). If the hue of the pixel is determined to be the allowable range obtained by the hue allowable range calculation processing 1005 for dealing with uneven color (1306), the pixel is classified according to the chroma difference between the pixel and the stamp color (1307) and returned (1311). If the saturation difference from the stamp color is lower than the threshold for dealing with uneven color, a character flag is set (1308). If the saturation difference from the stamp color is within a certain range of the threshold for dealing with uneven color, a character candidate flag is set (1309). If the saturation difference is larger than a certain range, an exclusion candidate flag is set (1310). If it is determined that the distance between the pixel and the stamp color is within the threshold or the hue is within the threshold (1312), the character candidate flag is set (1313) and returned (1314). In cases other than these determinations, a removal candidate flag is set (1315) for this pixel, and the process ends.

  FIG. 14 is a schematic diagram of binarization processing. FIG. 14A shows an example of binarization for the entire area of the date line. As a binarization method, many methods such as a method using a fixed threshold and a method of dynamically changing the threshold have been proposed. As a representative method, there is a binarization method of Otsu. In this method, a threshold for discriminating which of the two classes (whitened pixel or blackened pixel) each pixel belongs to a grayscale image is calculated using a discriminant analysis method. The threshold value is a value that minimizes the intra-class variance and maximizes the inter-class variance. When the date area 1402 of the grayscale image 1401 corrected by the inclination correction processing 507 is binarized by the binarization method of Otsu, a date area binary image 1403 is generated. However, in the binarization of Otsu for the entire date area with uneven color, there is a problem that a pixel with high luminance may become a whitened pixel depending on a threshold value. As a method for solving this problem, in the present invention, binarization is performed on a plurality of partial areas in the date area, and a binary image of the entire date area is obtained by logical sum of the binary images of the plurality of partial areas. The reason why the partial area is selected is that a pixel having a high luminance of the character component in the partial area has a high binarization threshold even when Otsu's binarization is used.

  FIG. 14B is an example of binarization using a plurality of partial area ORs according to the present invention. The partial area 1404 is an area having a specified size from the left end of the date area. When only this partial area is binarized using, for example, binarization of Otsu, a binary image binary image 1405 is obtained. The method of taking the partial area is not limited to this method. Also, the binarization method need not be limited to Otsu's binarization. Although the binary image 1406 becomes a binary image of the entire date area, here, it is the same as the binary image 1405 because only the partial area 1404 is binarized. The partial area 1407 is an area shifted to the X coordinate by ½ of the width of the partial area 1404. As a result of binarizing the partial area 1407 in the same manner, a binary image 1408 is obtained. The binary image of the entire date area is a binary image 1410 that is a logical sum (1409) of the entire date area binary image 1406 and the binary image 1408. Similarly, the partial area is binarized with respect to the partial area 1411 shifted to the X coordinate by ½. The shifting of the partial area 1412 is repeated until the right end of the date area is reached, and a binary image 1413 of the entire date area is obtained.

Character string recognition processing 405 is performed on the binary image obtained as described above.
<Second Embodiment>
Next, a second embodiment of a stamp date reading method and a stamp date reader to which the present invention is applied will be described. In the present embodiment, the dropout method can be designated as one for dealing with uneven color and for pre-printing / background pattern removal. The input image can be dealt with in any case by using a color unevenness countermeasure when the color unevenness is large, or using a preprint / background pattern removal dropout when there are many preprints or ground patterns.

  FIG. 15 is another diagram showing a processing flow of the second embodiment of reading the stamp date to which the color dropout corresponding to the color unevenness according to the present invention is applied. The processes with the same numbers as in FIG. 4 are the same as those in FIG. In this example, the color dropout method is designated in the color dropout method designation processing 1501. In the color dropout method determination 1502, if the method specified in the method specifying process 1501 is for color unevenness, color dropout processing 404 for outputting a color unevenness corresponding binary image is performed. In other cases, color dropout 1503 for outputting a pre-printed / background pattern removed binary image is performed. Note that the processing of 1501 is not limited to this position. A color dropout 1503 for outputting a pre-printed / background pattern removed binary image will be described with reference to FIG.

  FIG. 16 is a diagram showing a flow of a color dropout process for outputting a pre-print / background pattern removal binary image according to the present invention. The processes with the same numbers as in FIG. 5 are the same as those in FIG. In this color dropout process, pre-printing and a background pattern are correctly removed from the processing target area, a grayscale image in which only the stamp color is left is generated, and this is binarized to generate a binary image. The difference from FIG. 5 is that the color unevenness gray image generation process 506 is replaced with a pre-print / background pattern removal gray image generation process 1601. The pre-print / background pattern removal gray-scale image generation processing 1601 will be described with reference to FIG.

FIG. 17 is a diagram showing the flow of the pre-print / background pattern removal gray image generation process. The processes with the same numbers as those in FIG. 10 are the same as those in FIG. The pixel classification (for achromatic color) 1701 for pre-printing / background pattern removal for each pixel within the radius of the stamp will be described with reference to FIG. In the hue allowable threshold calculation 1702 for pre-printing / background pattern removal, for example, the allowable range 1105 in FIG. 11 is used. As an example of calculation of the allowable range for pre-printing / background pattern removal, the frequency distribution 1102 may be treated as a normal distribution centered on the hue 1101 and the allowable range may be a hue within a deviation βσ (for example, β = 1). In order to reduce the calculation amount, the allowable range may be from the frequency of the hue 1101 until the frequency becomes 1 / m (for example, m = 2) or less.
The pixel classification (for chromatic color) 1703 for pre-printing / background pattern removal will be described with reference to FIG.

  When the pre-printed ruled line or the like has a color similar to the stamp color, or when a false color occurs near the ruled line and becomes a color similar to the stamp color, it may remain as a character flag in the pixel classification 1701 or 1703. In response to this, ruled line removal 1704 is performed. The ruled line removal 1704 will be described with reference to FIG. The ruled line removal 1704 may be inserted between 1005 and 1006 in FIG.

  FIG. 18 is a determination flow of pixel classification for pre-printing / background pattern removal when the stamp color is an achromatic color. The same numbers as those in FIG. 12 are the same as those in FIG. The difference from FIG. 12 is that the determination (1207) is replaced with the determination (1801). Judgment (1801) is when the pixel saturation is equal to or lower than the slide saturation threshold of the pixel, the luminance difference from the stamp color is equal to or lower than the luminance difference threshold for pre-printing / background pattern removal, and the luminance is lower than the luminance of the stamp color. The character flag set (1205) and the process returns (1206).

  FIG. 19 is a determination flow of pixel classification for pre-printing / background pattern removal when the seal color is a chromatic color. The same numbers as those in FIG. 13 are the same as those in FIG. If it is determined that the distance from the stamp color is closer than the threshold (1901), the character flag is set (1902) and returned (1903). If it is determined that the hue difference from the stamp color is closer than the threshold and the luminance is lower than the luminance of the stamp color (1904), the character flag is set (1905) and returned (1906). When it is determined that the hue is within the allowable hue range for pre-printing / background pattern removal (1907), a determination based on the saturation difference (1908) is performed. In the determination (1908), if the saturation difference is lower than the pre-print / background pattern removal threshold, a character flag is set (1909). If the saturation difference is within a predetermined threshold range for pre-printing / background pattern removal, a character candidate flag is set (1910). Otherwise, the removal candidate flag is set (1911) and the process returns (1912).

  FIG. 20 shows an outline of ruled line removal. If the pre-printed ruled line is similar to the stamp color or a false color is generated near the ruled line due to the above pixel determination, these are set as a character flag and dropped out. May not be. In this case, there is a possibility that ruled line components may remain even after binarization, and the date line is mixed with ruled lines. In order to solve this, for example, filtering processing of 3 × 3 pixels is performed, and ruled line components are removed from thin vertical lines and horizontal lines.

  FIG. 20A shows an input image, and the date stamp 2001 is stamped on the ruled line 2002. When the ruled line 2002 is a color similar to the stamp color of the date stamp, the result of generating a gray image is as shown in FIG. 20B, and the binarization and date line extraction result of FIG. 2003 remains. As an example of a method for removing ruled lines, horizontal / vertical ruled line detection filtering processing 2004 is performed. If the flag of the pixel classified in the pixel classification 1701 or 1703 is a character flag, 3 × 3 pixels in the vicinity of 8 centering on this pixel are generated. In the flag of each pixel, the character flag 2005 and the flag of the pixel coordinates corresponding to the filter of the horizontal ruled line 2007 or the vertical ruled line 2008 of the 3 × 3 pixel set as the other flag 2006 are removed from the character flag or Convert to character candidate flag.

The result of performing the above horizontal / vertical ruled line detection filtering processing on all the pixels to generate a grayscale image is as shown in FIG. 20D, and the binarization and date line extraction result of FIG. The removed date character line.
<Third embodiment>
Next, a third embodiment of a stamp date reading method and a stamp date reader to which the present invention is applied will be described. In this embodiment, by generating dropout images for both color unevenness and pre-printing / background pattern removal in dropout, when color unevenness is large, a binary image for color unevenness is pre-printed. If there are many background patterns, a date image string can be read without depending on the input image because a correct date character line can be extracted from one of the binary images for pre-printing / background pattern removal.

  FIG. 21 is another diagram showing a processing flow of the third embodiment of reading a stamp date to which the color dropout corresponding to uneven color is applied according to the present invention. The processes with the same numbers as in FIG. 4 are the same as those in FIG. In a color dropout process 2101 for outputting two binary images for dealing with color unevenness and a binary image for preprinting / background pattern removal, a binary image for dealing with color unevenness and a binary image for preprinting / background pattern removal The two binary images are output. A color dropout process 2101 for outputting two binary images, one for dealing with uneven color and the other for pre-printing / background pattern removal, will be described with reference to FIG. In a date character string recognition process 2102 for the first binary image, date character string recognition is performed for the first binary image. In a date character string recognition process 2102 for the second binary image, date character string recognition is performed for the second binary image. A recognition result having a high score in the two date character string recognition results is set in 2104. There is no particular rule as to which of the binary image for color unevenness and the binary image for pre-printing / background pattern removal is the first.

FIG. 22 is a diagram showing a color drop-out processing flow for outputting two binary images, a non-uniformity binary image and a preprint / background pattern removal binary image. The processes with the same numbers as those in FIGS. 4 and 16 are the same as those in FIGS. Here, after performing common processing 501 to 505 for color unevenness correspondence and pre-printing / background pattern removal, first, binary images for color unevenness correspondence are generated (506 to 508). Next, binary images for pre-printing / background pattern removal are generated (1601, 507 to 508), and these two binary images are output.
<Fourth embodiment>
Next, a fourth embodiment of a stamp date reading method and a stamp date reader to which the present invention is applied will be described. In this embodiment, by generating dropout images for both color unevenness and pre-printing / background pattern removal in dropout, when color unevenness is large, a binary image for color unevenness is pre-printed. If there are many background patterns, a date image string can be read without depending on the input image because a correct date character line can be extracted from one of the binary images for pre-printing / background pattern removal. If the date character string reading with respect to the first binary image is successful, the reading of the date stamp of interest at that time is terminated, so that the processing time can be reduced as compared with the third embodiment.

FIG. 23 is another diagram showing a processing flow of the fourth embodiment of reading a stamp date to which the color dropout corresponding to uneven color according to the present invention is applied. The processes with the same numbers as those in FIGS. 4 and 21 are the same as those in FIGS. The difference from FIG. 21 is that the result of the date character string recognition processing 2102 for the first binary image after the date character string recognition processing 2102 for the first binary image has a recognition result score that is greater than or equal to the threshold value. If it is determined that there is (2301), the recognition result is set (2302), and the process is returned to the next date stamp (2303).
<Fifth embodiment>
Next, a fifth embodiment of a stamp date reading method and a stamp date reader to which the present invention is applied will be described. In the present embodiment, dropout is performed separately for color unevenness handling and preprinting / background pattern removal, and date character string recognition is performed after each dropout.

FIG. 24 is another diagram showing a processing flow of the fifth embodiment of reading the stamp date to which the color dropout corresponding to uneven color is applied according to the present invention. The processes with the same numbers as those in FIGS. 4, 15, and 21 are the same as those in FIGS. Here, a color dropout process 404 for generating a binary image for color unevenness is performed, and a date character string recognition process 2102 is performed using this as a first binary image, and then for pre-printing / background pattern removal. A color dropout process 1503 for outputting a binary image is performed, a date character string recognition process 2103 is performed using this as a second binary image, and a recognition result having a high score is set (2104). Note that 404 and 1503 in FIG. 24 may be interchanged.
<Sixth embodiment>
Next, a sixth embodiment of a stamp date reading method and a stamp date reader to which the present invention is applied will be described. In this embodiment, dropout is performed separately for color unevenness handling and pre-printing / background pattern removal, and the date character string is read after each dropout. When the result of date string recognition by the first dropout is obtained, the reading of the date stamp of interest at that time is terminated, so that the color drop processing time and date characters are compared with the fifth embodiment. Processing time for column recognition can be reduced.

FIG. 25 is another diagram showing a processing flow of the sixth embodiment of reading the stamp date to which the color dropout corresponding to uneven color is applied according to the present invention. The processes with the same numbers as those in FIGS. 4, 15, 21, and 23 are the same as those in FIGS. 4, 15, 21, and 23. The difference from FIG. 24 is that the result of the date character string recognition processing 2102 for the first binary image after the date character string recognition processing 2102 for the first binary image has a recognition result score that is greater than or equal to the threshold value. If it is determined that there is (2301), the recognition result is set (2302), and the process is returned to the next date stamp (2303).
<Seventh embodiment>
Next, a seventh embodiment of a stamp date reading method and a stamp date reader to which the present invention is applied will be described. In this embodiment, it is expected that there is a binary image that can be correctly recognized by performing color dropout with various threshold values, generating a plurality of binary images based on the threshold, and recognizing the date character string.

FIG. 26 is another diagram showing a processing flow of the seventh embodiment of reading a stamp date to which a color dropout for outputting a plurality of binary images is applied. The processes with the same numbers as those in FIGS. 4 and 23 are the same as those in FIGS. A color drop-out process 2601 for outputting a plurality of binary images generates binary images in which various threshold values shown in FIGS. 5 to 19 are changed stepwise, and the number of binary images (2602) includes date characters. Column recognition processing 405 is performed, and if the recognition result score is determined to be above (2301), the recognition result is set (2302) and the processing returns to the next date stamp (2303).
<Eighth embodiment>
Next, an eighth embodiment of a stamp date reading method and stamp date reader to which the present invention is applied will be described. In this embodiment, color dropout is performed with various threshold values, a plurality of binary images are generated based on the threshold values, date character strings are recognized for all the binary images, and a result with the highest accuracy can be obtained. it can.

  FIG. 27 is another diagram showing a processing flow of the eighth embodiment of reading a stamp date to which a color dropout for outputting a plurality of binary images is applied. The processes with the same numbers as those in FIGS. 4 and 26 are the same as those in FIGS. The difference from FIG. 26 is that, after date character string recognition 405 is performed on all binary images in 2602, the recognition result having the highest score among all the recognition results is set (2701). Further, instead of 2701, the majority of date character strings recognized among all the recognition results may be taken to set the most recognition results.

  Next, a display screen for confirming and correcting the recognition result by the stamp date reading method and the stamp date reader to which the present invention is applied will be described. The contents to be displayed are not limited by the following description.

  FIG. 28 is a display example of a screen for confirming and correcting the result of the date stamp reading process of FIGS. 21, 23, 24, and 25. The display device 50 displays a screen 2801. This is a window for displaying the result / confirmation of the date stamp reading process in 2802. The input image display window 2803 displays a color image 2804. A processing target area 2806 of the date stamp reading process including the date stamp 2805 detected by the date stamp detection is displayed. The form ID is displayed in 2807. A number 2808 is displayed to indicate the number of the color image currently being confirmed / corrected among all the processed color images. 2809 displays grayscale image generation results in the color dropout processing for color unevenness and pre-printing / background pattern removal according to the present invention. A binary image obtained as a result of binarization and date row extraction is displayed in 2810 for color unevenness correspondence and pre-printing / background pattern removal. A date character string as a result of the date reading is displayed in 2812. Reference numeral 2813 denotes an editor for correcting the date reading result. Reference numerals 2814 and 2815 denote buttons for switching to display of the recognition result of another date stamp if there is another date stamp in the same color image. Reference numeral 2816 denotes a scroll bar for switching to display of a date stamp recognition result for another color image.

  FIG. 29 is another display example of a screen for confirming and correcting the result of the date stamp reading process of FIG. 4 or FIG. This display confirms whether the reading result is obtained when various threshold values such as color dropout are changed when the date stamp reading result is a non-reading result. Alternatively, by automatically varying various threshold values such as color dropout and adjusting until a reading result is obtained, correction when the date stamp reading result is a non-reading result can be automatically performed. The display device 50 displays a screen 2901. This is a window for displaying 2902 the result / confirmation of the date stamp reading process. An input image display window 2903 displays a color image 2904. A processing target area 2906 of the date stamp reading process including the date stamp 2905 detected by the date stamp detection is displayed. The form ID is displayed in 2907. A number 2908 is displayed to indicate the number of the color image currently being confirmed / corrected in all the processed color images. Reference numeral 2909 denotes a date stamp area image obtained by extracting the processing target area 2906 from the color image. 2910 displays a grayscale image generation result in the color dropout processing in the processing of FIG. 4 or FIG. 15, a binary image resulting from binarization and date row extraction, and a date character string as a result of date reading. . 2911 is a title of 2912 and 2913. 2912 changes various threshold values in the color dropout processing manually or automatically when the recognition result of 2910 is unread, etc., and performs the processing of FIG. 4 or FIG. Adjust what you get. 2913 displays the processing result of FIG. 4 or FIG. 5 by the adjustment change of the threshold adjustment of 2912. FIG. Reference numerals 2914 and 2915 are buttons for switching to the display of the recognition result of another date stamp if there is another date stamp in the same color image. Reference numeral 2916 denotes a scroll bar for switching to display of a date stamp recognition result for another color image.

  FIG. 30 is another display example of a screen for confirming and correcting the result of the date stamp reading process of FIG. 26 or 27. The display device 50 displays a screen 3001. This is a window for displaying the result / confirmation of the date stamp reading process at 3002. The input image display window 3003 displays a color image 2904. The processing target area 3006 of the date stamp reading process including the date stamp 3005 detected by the date stamp detection is displayed. The form ID is displayed at 3007. A number indicating the number of the color image currently being confirmed / corrected in all the processed color images is displayed at 3008. Reference numeral 3009 denotes a plurality of grayscale image generation results in the color dropout processing obtained from the color dropout processing results of FIG. 26 or FIG. 27, or binary images resulting from binarization and date row extraction, or both Are displayed in whole or in part. Reference numeral 3010 displays a highlight frame for an image accepted as a result of date reading. A window 3011 displays parameter values such as color dropout for an image selected by an input device or the like. Reference numeral 3012 denotes an editor for correcting the date reading result. Reference numerals 3013 and 3014 denote buttons for switching to display of a recognition result or the like of another date stamp if another date stamp exists in the same color image. Reference numeral 3015 denotes a scroll bar for switching to display of a date stamp recognition result for another color image.

  Subsequently, characteristic configurations of the network control system obtained by the above-described embodiments will be listed below.

(1) A method for reading a stamp date from a color image stamped with a date stamp,
Inputting the color image;
Detecting a date stamp from the color image;
A step to specify the dropout method;
Determining whether the specified dropout method is for color unevenness or pre-printing / background pattern removal;
The image processing method according to claim 1, wherein if the specified dropout method is for color unevenness, a date character line binary image is output for the detected date stamp;
3. The image processing method according to claim 2, wherein a date character line binary image is output for the detected date stamp if the specified dropout method is other than for color unevenness, and a date character string for the binary image. The step of recognizing
Setting the recognition result;
A second stamp date reading method comprising the step of outputting a result.

(2) A dropout image processing method that leaves only character components from the processing target image,
Inputting the processing target image;
Color clustering the pixels of the image to be processed, selecting a background color based on the color clustering,
Selecting a stamp color from the pixels of the processing target image;
Color space conversion of pixels of the processing target image;
Generating a gray image corresponding to color unevenness from pixels of the processing target image;
Correcting the gradient of the grayscale image for color unevenness,
Binarizing a grayscale image obtained by correcting inclination of the grayscale image for color unevenness;
Generating a gray image for pre-printing / background pattern removal from pixels of the processing target image;
Tilt correcting the gray image for pre-printing / background pattern removal;
A third image processing method comprising a step of binarizing the grayscale image obtained by correcting the inclination of the grayscale image for pre-printing / background pattern removal.

(3) A method of reading a stamp date from a color image on which a date stamp is stamped,
Inputting the color image;
Detecting a date stamp from the color image;
The image processing method according to claim 16, wherein two binary images, a binary image for color unevenness and a binary image for preprint removal, are output for the detected date stamp;
Recognizing a date string for the first binary image;
Recognizing a date character string for the second binary image;
Setting a recognition result having a higher score of the above two recognition results;
A third imprint date reading method comprising a step of outputting a result.

(4) A method for reading a stamp date from a color image on which a date stamp is stamped,
Inputting the color image;
Detecting a date stamp from the color image;
The image processing method according to claim 16, wherein two binary images, a binary image for color unevenness and a binary image for pre-printing / background pattern removal, are output for the detected date stamp;
Recognizing a date character string for the first binary image;
Determining a score of a date string result for the first binary image;
A step of setting a recognition result when the score is equal to or greater than a threshold in the above determination;
A step of moving the processing to the next date stamp after setting the recognition result when the score is equal to or greater than a threshold in the above determination;
Recognizing a date character string for the second binary image;
Setting a recognition result having a higher score of the above two recognition results;
A fourth stamp date reading method comprising a step of outputting a result.

(5) A method for reading a stamp date from a color image on which a date stamp is stamped,
Inputting the color image;
Detecting a date stamp from the color image;
2. An image processing method according to claim 1, wherein a binary image for color unevenness is output for the detected date stamp;
A step of recognizing a date character string for the first binary image and outputting a binary image for preprint / background pattern removal for the detected date stamp for the output binary image Item 10 image processing method;
A step of recognizing a date character string with respect to a second binary image with respect to the output binary image, and a step of setting a recognition result having a higher score of the two recognition results;
A fifth stamp date reading method comprising a step of outputting a result.

  (6) The stamp date reading method according to claim 19, characterized in that the order of dropout described in (5) is not the same for color unevenness handling and for pre-printing / background pattern removal.

(7) A method of reading a stamp date from a color image on which a date stamp is stamped,
Inputting the color image;
Detecting a date stamp from the color image;
2. An image processing method according to claim 1, wherein a binary image for color unevenness is output for the detected date stamp;
Recognizing a date character string for the first binary image and determining a date character string result score for the first binary image for the output binary image;
A step of setting a recognition result when the score is equal to or greater than a threshold in the above determination;
A step of moving the processing to the next date stamp after setting the recognition result when the score is equal to or greater than a threshold in the above determination;
The image processing method according to claim 10, wherein a binary image for pre-printing / background pattern removal is output for the detected date stamp;
A step of recognizing a date character string with respect to a second binary image with respect to the output binary image, and a step of setting a recognition result having a higher score of the two recognition results;
A sixth imprint date reading method comprising a step of outputting a result.

  (7) The stamp date reading method according to claim 21, characterized in that the order of dropouts described in (6) is not the same for color unevenness correspondence and preprint / background pattern removal.

(8) A method for reading a stamp date from a color image on which a date stamp is stamped,
Inputting the color image;
Detecting a date stamp from the color image;
Performing a color dropout for outputting a plurality of binary images to the detected date stamp;
Recognizing a date character string for the output binary image and determining a score of a date character string recognition result for the binary image;
A step of setting a recognition result when the score is equal to or greater than a threshold in the above determination;
A step of moving the processing to the next date stamp after setting the recognition result when the score is equal to or greater than a threshold in the above determination;
A seventh imprint date reading method comprising a step of outputting a result.

(9) A method for reading a stamp date from a color image on which a date stamp is stamped,
Inputting the color image;
Detecting a date stamp from the color image;
Performing a color dropout for outputting a plurality of binary images to the detected date stamp;
A step of recognizing a date character string for all the output binary images and a step of setting a recognition result having the highest score among the results of date character string recognition for all the binary images;
An eighth stamp date reading method comprising a step of outputting a result.

  (10) The stamp according to claim 24, wherein in the step of setting the recognition result according to (9), the frequency of the character string of the recognition result is counted, and the most frequently recognized character string can be set as the recognition result. Date reading method.

DESCRIPTION OF SYMBOLS 10 ... Input device, 20 ... Image input device, 30 ... CPU, 40 ... Dictionary, 50 ... Display device, 60 ... Image DB, 210 ... Color dropout binary image for uneven color, 404 ... Binary for uneven color Color dropout processing for outputting an image, 405, date character string recognition processing, 502... Color clustering processing, 503... Background color selection processing, 504, Stamp color selection processing, 506. Inclination correction processing, 508..., Binarization processing, 1503... Color dropout processing for outputting a pre-print / background pattern removal binary image, 1601... Pre-print / background pattern removal grayscale image generation processing, 2101. Color dropout processing that outputs two binary images: a binary image and a binary image for pre-printing / background pattern removal

Claims (7)

A dropout image processing method that leaves only character components from a processing target image,
Inputting the processing target image;
Color clustering the pixels of the image to be processed, selecting a background color based on the color clustering,
Selecting a stamp color from the pixels of the processing target image;
Color space conversion of pixels of the processing target image;
Generating a gray image corresponding to color unevenness from pixels of the processing target image;
Correcting the gradient of the grayscale image;
A first image processing method including a step of binarizing the grayscale image subjected to inclination correction ,
In the step of selecting the stamp color,
Means for selecting a circumferential color of pixels near the circumference of the imprint of the image to be processed;
An image processing method comprising: selecting a stamp color by means for determining a color closest to the selected circumferential color from pixels in a stamp area of the processing target image. In the gray image generation step for dealing with color unevenness,
Means for converting the stamp color into a color space;
Means for determining from the saturation and luminance of the stamp color obtained by the color space conversion to an achromatic or chromatic color;
If the determination result is an achromatic color, means for performing pixel classification for achromatic unevenness for achromatic color,
If the determination result is a chromatic color, means for calculating a hue tolerance range for color unevenness,
Means for performing pixel classification for color unevenness for chromatic colors;
The image processing method according to claim 1, further comprising means for converting the result of pixel classification into a gray value. In the pixel classification means for dealing with uneven color for the achromatic color,
The image processing method according to claim 2, wherein each pixel in the processing target image is discriminated into a background, a character, and a removal candidate using the selected stamp color, background color, saturation and luminance of the pixel. In the hue tolerance calculation means for dealing with the color unevenness,
The frequency distribution is centered on the hue of the stamp color using the hue of the stamp color converted by the color space according to claim 2 and the frequency distribution of the hue of the pixel of the image to be processed by the color space conversion according to claim 1. The image processing method according to claim 2, wherein the image processing method is treated as a normal distribution, and a hue range up to a deviation ασ is set as a hue tolerance range for color unevenness. In the hue tolerance calculation means for dealing with the color unevenness,
Using the hue of the stamp color after the color space conversion according to claim 2 and the frequency distribution of the hue of the pixel of the image to be processed by the color space conversion according to claim 1, the frequency of the hue of the stamp color in the frequency distribution is determined. The image processing method according to claim 2, wherein a hue range up to a threshold value or less is set as a hue tolerance range for uneven color. In the pixel classification means for color unevenness for the chromatic color,
The image according to claim 2, wherein each pixel in the processing target image is discriminated into a background, a character, a character candidate, and a removal candidate using the selected stamp color, background color, hue and saturation of the pixel. Processing method. In the binarization step,
In the binarization, a partial area of the date area in the grayscale image subjected to the inclination correction is binarized according to a binarization threshold value of each partial area to generate a binary image, and a logical sum of each partial area binary image is calculated. The image processing method according to claim 1, wherein a binary image of a date area is obtained by taking the image.