JP5887242B2 - Image processing apparatus, image processing method, and program - Google Patents

Description

  The present invention relates to an image processing technique such as an OCR (Optical Character Reader), and more particularly to a binarization technique that separates a character and a background even when the character and the background have the same color and low contrast.

  A binarization process for generating a binary image from a color image is a basic process as a pre-process of OCR. Forms that are the main reading targets of OCR include preprinted ruled lines and characters called pre-printing. When pre-printing is close to a character to be read, character recognition may be an obstacle. Therefore, in the binarization processing in the conventional form OCR, binarization is generally performed in which the written characters are black (blackening) and the pre-printing and the background are white (whitening). As a conventional technique for binarization processing using a grayscale image as an input, a technique for obtaining a binarization threshold value of a luminance value using the Otsu method, Niblack method, or the like is known.

  As a prior art of binarization processing using a color image as input, Patent Document 1 discloses a luminance of pixels other than written characters such as pre-printing among three colors of RBG (red, blue, and green) that are captured by a color image. Describes a method of selecting a color with the highest value and then binarizing it by determining a threshold value from the gray value of the pixel in that color. Patent Document 2 describes a method of separating a character color and a pre-print color by extracting only the character color after estimating the character color. Japanese Patent Application Laid-Open No. 2004-133830 defines a binarization target region using layout analysis processing, and extracts background and ruled line colors with reference to format information in which a pre-printing position or the like is defined in the region. Describes a binarization method for separating pre-printing and background and characters.

  On the other hand, in the form recognition technology, which is the latter process of binarization, it is a problem to specify the attributes of characters to be recognized (eg, amount, date, name) and recognize the written characters with high accuracy. is there. Conventionally, as described in Patent Document 3, it is common to cut out and recognize a character area to be recognized using format information of a form in which character positions and attributes are defined in advance. On the other hand, in Patent Document 4, there is no predefined format information by recognizing characters on the entire form and extracting pairs of item names (for example, money) and data (for example, 10,000 yen) adjacent to each other. Describes the technology for recognizing the attributes of written characters.

JP 2003-196292 A JP 2010-244372 A JP 2005-258683 A JP 2011-248609 A

  In the technique of Patent Document 4, since it is necessary to recognize an item name, it is necessary to recognize a pre-printed character. Therefore, in addition to the written characters, the preprinted characters need to be blackened correctly. However, pre-printed characters on general forms are often printed in a color similar to the background color and with small shade differences and color differences. This is because the purpose is to whiten simultaneously with the background color in the binarization process. Therefore, it is difficult to correctly blacken the preprinted characters by the binarization processing of the methods of Patent Documents 1 to 3.

  As an example of an object to be read by the form recognition technique of Patent Document 4, a problem of binarization of pre-printing will be described using a transfer slip. The transfer slip includes pre-printed ruled lines, pre-printed characters, and the like, and printed characters and handwritten characters are described. The written characters, pre-printing, and paper background color (background color) are different for each type of slip. With such a form, it is difficult to obtain an appropriate binary image by a normal binarization method. Specifically, if the threshold value for binarization is lowered in order to blacken a light-colored character or preprint, both the character portion and the background become black in a dark background color form. On the other hand, if the threshold is increased to whiten a dark background color, light characters and preprints are also whitened. Furthermore, it becomes more difficult when a plurality of forms having different shade values of characters and backgrounds are mixed.

  In the method described in Patent Document 1, it is assumed that characters and pre-printing are different colors. Therefore, when characters and pre-printing are similar colors, pre-printing is whitened. In the method described in Patent Document 2, since the color other than the estimated character color is whitened, pre-printing is whitened. The method described in Patent Document 3 requires pre-defined form format information, and therefore cannot cope with a mixture of various forms.

  In view of the problems in the prior art as described above, the object of the present invention is to properly blacken characters and preprints regardless of the size of the written characters and preprints and the contrast of the background color. It is to binarize and recognize characters.

  In order to solve the above problems, the present invention is an image processing apparatus that generates a binary image from a color image obtained by imaging with a scanner, and for generating the color image and the binary image. Analyzing the layout of the color image, a storage unit that stores predetermined determination criteria, and using the result of the layout analysis, a representative color of a background color, a preprint color, and a character color is obtained by color clustering, and a character color and Calculate the background color separation standard, the pre-print color and background color separation standard, the character color and the pre-print color separation standard, and the previous color separation based on the color information and layout information of each pixel in the color image An image processing unit that performs binarization processing by switching a reference and stores the binarized pixel value in the storage unit.

  In the binarization according to the present invention, it is possible to correctly extract written characters and pre-prints regardless of the size of the written characters and pre-prints, and the contrast of the background color. As a result, both written characters and pre-printed characters can be recognized with high accuracy.

It is a block diagram of the image processing apparatus of 1st Embodiment. It is an example of a form image to be input. 2 is an example of a binary image targeted by the present invention. It is a figure which shows the example of the binary image by a prior art. It is a figure which shows the example of the binary image by a prior art. It is a figure which shows the outline of the binarization process of embodiment of this invention. It is a figure which shows the outline of the layout analysis process of embodiment of this invention. It is a figure which shows the example of a color clustering. It is a figure which shows the outline of HSV color space. It is a figure which shows the example of the input image and the binary image by this invention. It is a figure which shows the outline of the binarization process of 2nd Embodiment of this invention. FIG. 3 is a diagram showing an example in which a background color is detected from the input image of FIG. It is a figure which shows the outline of the binarization process of 3rd Embodiment of this invention. It is a figure which shows the outline of the binarization process of 4th Embodiment of this invention. It is a figure which shows the outline of the binarization process of 5th Embodiment of this invention. It is a figure which shows the outline of the binarization process of 6th Embodiment of this invention. It is a figure which shows the relationship between the brightness | luminance of a background color, and a separation standard. It is a figure which shows the outline of the process which switches and separates a separation reference | standard in 6th Embodiment of this invention. It is a figure which shows the outline of pixel classification in 6th Embodiment of this invention. It is a figure which shows the outline of the rebinarization process based on the classification result in 6th Embodiment of this invention. It is a figure which shows the example of the rebinarization which blackens a description character. It is a figure which shows the example of the rebinarization which blackens preprinting. It is a figure which shows the example of the rebinarization which whitens noise. It is a block diagram of the image processing apparatus which performs binarization and character recognition. It is a block diagram of a sales office window terminal.

  DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, modes for carrying out the present invention (hereinafter referred to as embodiments) will be described by dividing them into first to eighth embodiments with reference to the drawings (refer to drawings other than the referenced drawings as appropriate). To do.

(First embodiment)
First, the configuration of the image processing apparatus according to the first embodiment will be described. As illustrated in FIG. 1, the image processing apparatus 100 is a computer apparatus that generates a binary image from a color image acquired by capturing a form, a document, or the like, and includes an operation unit 110, a display unit 120, and an imaging unit 130. And a storage unit 140 and a control unit 150.

  The operation unit 110 is an operation unit including a keyboard used when a user performs input operations of numbers and characters, a button for instructing start of scanning, and the like. The display unit 120 is a display unit including a liquid crystal display that displays an input image, a binary image, and the like. The imaging unit 130 is an imaging unit such as a digital camera that captures an image using a flatbed scanner, a scanner with an auto sheet feeder, and an imaging device such as a CCD (Charge Coupled Device).

  The storage unit 140 is a storage device including a memory, a hard disk device, and the like, and stores images captured by the imaging unit 130, various internal processing programs, and various setting data. In addition, for example, the storage unit 140 selects a color with the highest luminance among the obtained representative colors of each cluster as a predetermined determination criterion for detecting the background color of the subject (such as a document) from a plurality of representative colors. The reference for the background color is stored. In addition, the storage unit 140 stores setting values for binarization, determination criteria, and the like.

  The control unit 150 is a control unit that performs image processing such as image capturing processing and input image binarization processing, and includes an image capturing processing unit 151 and an image processing unit 152.

  The imaging processing unit 151 is a control unit that performs imaging control of a subject using the imaging unit 130. The image processing unit 151 develops and temporarily stores image data as an input image on a memory, and stores the image data as a file. 140 can be stored.

  The image processing unit 152 is a unit that removes the background from the input image created by the imaging processing unit 151 and binarizes it, and is a characteristic part of the image processing apparatus 100 of the first embodiment. Specifically, the image processing unit 152 first performs layout analysis on the image created and input by the imaging processing unit 151 to obtain a background color, a pre-print color, and a character color. Next, a standard for separating the background color and the pre-print color, the background color and the character color, and the pre-print color and the character color is calculated. Finally, using the layout information and color information, a binary image is generated by switching three types of color separation criteria for each pixel in the image. This processing flow will be described later with reference to FIG.

  Before showing a specific example of the binarization method realized in the present invention, an outline of the conventional binarization processing result and the processing result of the present invention will be described with reference to FIGS.

  FIG. 2 shows an input image to be binarized. On a white paper form 200, there are a white background color area surrounded by a ruled line 210, a dark background color area surrounded by a dark ruled line 220, and a light background color area surrounded by a thin ruled line 230, respectively. In this area, pre-printed characters 240 to 260 on which “amount” is written exist. Further, characters to be recognized are written in a frame such as 270. The colors of the pre-printed characters 250 and 260 are similar to the background color and have a slightly lower luminance than the background color. An object of the present invention is to generate a binary image from the image in FIG. 2 in which the written characters, pre-printed characters, and ruled lines are blackened and the other background regions are whitened as shown in FIG. .

  The result of processing the image of FIG. 2 by the conventional binarization method is shown in FIGS. FIG. 4 shows the result of binarizing the entire form 200 with a low threshold for the purpose of blackening the thin preprint. Although the ruled line 430 and the pre-printed characters 460 can be blackened, the dark background region including the pre-printed characters 450 is all blackened. On the other hand, FIG. 5 shows the result of binarizing the entire form 200 with a high threshold for the purpose of whitening a dark background region. Since the ruled line 530 and the preprinted character 560 are equal to or greater than the threshold value, they are whitened. 4 and 5, it can be seen that if the entire form is binarized on a uniform basis, it is difficult to recognize both the written characters and the pre-printed characters with high accuracy. In order to deal with this problem, there is a conventional method in which a threshold value is calculated for each local region and binarized. However, if the hue and brightness of the background color and the pre-printed character are close as shown in FIG. 2, noise caused by blackening a part of the background or blurring of a character caused by whitening a part of the character. Etc. occur frequently. Thus, it has been difficult to generate a binarized image that extracts both written characters and pre-printed characters from a form that has a large shade difference between the background color and pre-print. This is the same even when a form with thin pre-printed characters and a dark form are mixed. Details of the solution to the problem in the present invention will be described in FIG.

  An example of the binarization processing flow executed by the image processing unit 152 in the present invention is shown in FIG. The contents of the processing in each step are as follows. Note that the following description is based on the premise that a color image is input, but it can also be applied to gray images by using pixel gray values (luminance values, brightness values).

  In step 600 of FIG. 6, layout analysis is performed to detect the positions of ruled lines, frames, and character lines on the form.

  Details of the layout analysis processing executed by the image processing unit 152 are shown in FIG. First, in step 700, the form image is binarized. In this binarization processing, not only the pre-printed characters and the written characters are blackened as shown in FIG. 3, but at least the ruled lines are blackened. The threshold value can be set more easily than when the binarized image of FIG. 3 is generated. This point will be described in detail when the frame is extracted in step 720. In step 710, ruled lines are extracted from the binarized image generated in step 700. As an example of the ruled line extraction method, there is a method of extracting a vertical ruled line and a horizontal ruled line from the continuity of black pixels in the vertical direction and the horizontal direction. In step 720, a frame is extracted from the extracted ruled line. As an example of the frame extraction method, there is a method described in Japanese Patent Laid-Open No. 11-53466 in which a frame is extracted using features near the intersection of ruled lines. With this method, a frame can be extracted with high accuracy even if blurring or noise occurs on a ruled line. Since it is not necessary to accurately extract ruled lines, it is easy to set a threshold value for binarization processing in step 700. However, if it is determined that a frame cannot be extracted with a predetermined binarization threshold, it is possible to return to step 700 and change the binarization threshold and then perform the processing from step 710 again. Finally, in step 730, a character line is extracted from the binarized image generated in step 700. As a conventional method of this processing, there is a method called a connected component fusion method. In this method, a region where black pixels are connected is used as a connected component, and a character line is detected from the vertical / horizontal alignment of the connected component. Since the purpose is to detect the approximate position of a character line, a binary image in which the character is blurred or noise may be used. This also shows that the threshold setting in step 700 is easy. Note that when extracting ruled lines and character pixels in step 600, instead of the method shown in FIG. 7, a filter for detecting the contour such as a Laplacian filter may be used after converting the input color image into a gray image. The Laplacian filter is a filter that extracts a portion where the amount of change in luminance difference is extremely large. The coefficient of the 3 × 3 filter is

Given in.

  In step 610 of FIG. 6, the background color of the form is obtained. As an example of the background color extraction method, there is a method described in Japanese Patent Laid-Open No. 2012-95002. The background color detection described in this method divides a form image into a plurality of partial areas, maps each pixel in the partial area onto a color space such as RGB, and performs clustering, and stores it from the obtained clusters. This is a process of selecting a background color cluster using a predetermined criterion stored in the section 140 and setting the representative color of this cluster as the background color of the partial area.

  FIG. 8 is an example showing an RGB color space. The RGB color space is a three-dimensional space having R (red), G (green), and B (blue) values as axes, and the values of each RGB axis range from 0 to 255. The origin (0, 0, 0) represents black and (255, 255, 255) represents white. Each pixel in the processing target area can be mapped to one point on the RGB color space. When the pixels in the region are mapped, a plurality of distributions centered on representative colors are obtained. FIG. 8 shows an example in which red pre-printing (810) and green pre-printing (820) are printed on a white background color (800), and an area where black characters (840) are described is color clustered. . Color clustering is a process of clustering each pixel in a processing area on a color space. Specifically, this is realized by dividing the color space into regions using the Voronoi division, the Euclidean distance threshold value, and the like as the same cluster. In each cluster, the color located at the center of the cluster or the color with the highest frequency is set as the representative color of the cluster. To select the background color cluster (800) from the clusters in the space, criteria such as the highest brightness of the representative color of the cluster, the highest brightness, and the highest frequency of the cluster (wide area) should be selected. Use. The lightness will be described later in character color detection in step 630.

  Note that color clustering may use all of the pixels in the region or some of the pixels. For example, pixels sampled at regular intervals may be used for color clustering. Furthermore, in this example, pixel values are directly subject to clustering, but in order to reduce the effects of false colors and noise that occur due to CCD characteristics, average values with weighted pixels and weighted average values are set. This value may be obtained and used for color clustering.

  In step 620 of FIG. 6, the color of the ruled line of the form image is obtained. This process can be obtained by color clustering the pixels extracted as ruled lines in the ruled line extraction 710. When a cluster having a color different from the background color is obtained as a result of the color clustering, the representative color of this cluster is set as the ruled line color. There may be a plurality of ruled line colors. Generally, the color of the ruled line on the form and the color of the pre-printed character are often the same, and thus the color of the ruled line obtained here can be regarded as the pre-printed color. Note that the pixel to be processed in step 620 may be a pixel of a ruled line determined as a frame ruled line in the frame extraction process 720, instead of a pixel extracted as a ruled line in the ruled line extraction 710. Pixels extracted as ruled lines contain noise components of the same color as the ruled lines, but the ruled lines that make up the extracted frame have a low probability of containing noise, so the color of the ruled lines is obtained more accurately. be able to.

  In step 630, the color of the character is obtained. This process can be obtained by color clustering the pixels in the area extracted as a character line by the character line extraction 730. When a cluster having a color different from the background color is obtained as a result of the color clustering, the representative color of this cluster is set as the character color. There may be a plurality of character colors. As an example of a criterion for selecting a character color cluster, the cluster representative color has the lowest luminance, the lowest brightness, the lowest saturation, and a hue different from the background or pre-printing.

  Here, the HSV color space that expresses colors by using the indices of hue (H), saturation (S), and brightness (V) will be described with reference to FIG.

The RGB color space is easy to handle with a computer because colors are expressed using three colors of RGB. On the other hand, it is known that humans perceive colors not by RGB values but by indices of hue (Hue: color type), saturation (Saturation: vividness), and lightness (Value: brightness). Therefore, in the color clustering and color discrimination processing of the present invention, not only the RGB color space but also an index of the HSV color space close to humans may be used. The HSV color space can be visualized as a downward cone as shown in FIG. Details of each index of HSV are as follows.
Hue (H): Color type (red, blue, yellow, green, etc.). 0 to 360 ° range (changes on the circumference)
Saturation (S): Vividness of color. 0.0 to 1.0 (or 0 to 255) range (changes on the horizontal axis)
Lightness (V): The brightness of the color. Range of 0.0 to 1.0 (or 0 to 255) (changes on the vertical axis)
Conversion from the RGB color space to the HSV color space can be performed using Equations 2 to 5. Experiments have shown that human perception is most sensitive to changes in hue compared to changes in lightness and saturation. An HSL color space or the like may be used instead of the HSV color space.

  In step 640 of FIG. 6, a color separation criterion for binarization is obtained. Step 640 includes three steps 650 to 670.

  In step 650, a reference for separating the character color and the background color is obtained based on the information on the background color obtained in step 610 and the information on the character color obtained in step 630. As a separation standard for the character color and the background color, there are a method for obtaining a plane or a curved surface for separating clusters in the RGB color space and the HSV color space, and a method for obtaining a threshold value from luminance and lightness histograms. If at least one of the colors is a chromatic color having a certain degree of saturation (such as white to gray to black, which is not near the V-axis in FIG. 9), by limiting the hue and saturation range, There is a method of separation. These may be used alone or in combination. In general, for discrimination between chromatic colors, separation is performed with the highest priority on hue, and when hues are close, separation is performed using lightness or saturation.

  In step 660, the separation standard for the pre-print color and the background color is obtained, and in step 670, the separation standard for the character color and the pre-print color is obtained in the same manner as in step 650. In the present invention, “character color” indicates the color of a written character to be read, and “pre-print color” indicates the color of a ruled line or pre-print character. Although both presuppose that it is different, they may be the same. Even in this case, it is possible to blacken the written characters and the preprint and whiten the background. Depending on the application, one or two of the processes 650 to 670 may be used.

  In step 680 of FIG. 6, RGB information and HSV values of pixels, background information obtained in steps 610, 620, and 630, color information such as pre-printing, character RGB values and HSV values, and the layout obtained in step 600. Using the information, binarization is performed by switching the binarization separation criterion obtained in step 640 for each pixel. In this processing, the reference used differs depending on the type of binary image to be generated.

  An example of three types of binary images will be described with reference to FIG. FIG. 10A shows an input image in which preprinted characters and ruled lines of the same color as the background color are printed in a region with a background color, and numbers with low luminance are described as recognition targets.

  FIG. 10B is a binary image obtained by whitening the background of the image of FIG. 10A and blackening the written characters, pre-printed characters, and ruled lines. This binary image is generated when the purpose is to recognize both preprinted characters and written characters. In order to generate this binary image, both the character color-background color separation criterion in step 650 and the pre-print color-background color separation criterion in step 660 are used. As a specific example of the use of the separation criterion, there is a method of blackening only pixels that are determined not to be the background according to at least one of the separation criteria.

  FIG. 10C is a binary image obtained by whitening the background, preprinted characters, and preprinted ruled lines, and blackening the written characters from the image of FIG. This binary image is generated when the purpose is to reduce the adverse effects of pre-printing when recognizing written characters. This is particularly effective when preprinting and written characters are in contact. In order to generate this binary image, both the character color-background color separation criterion in step 650 and the character color-pre-print color separation criterion in step 670 are used. As a specific example of the use of the separation standard, there is a method of blackening a pixel determined to be a character color by both separation standards.

  FIG. 10D is a binary image obtained by whitening the written characters and the background and blackening the pre-printed characters and the ruled lines from the image of FIG. Generated for the purpose of recognizing only pre-printed characters. In order to generate this binary image, both the pre-print color-background color separation criterion in step 660 and the character color-pre-print color separation criterion in step 670 are used. As a specific example of the use of the separation standard, there is a method of blackening pixels determined to be pre-printed colors by both separation standards.

  For example, the user selects whether to output in one of FIGS. 10B, 10C, or 10D from the operation unit or the like, and the image processing unit outputs in the format specified by the operation unit or the like. The processing of step 680 is executed so as to do this.

  In step 680, not only can the separation criterion be switched based on the color of the pixel, but the separation criterion can be changed based on the coordinates of the pixel and the layout analysis results. This is because by using the layout analysis result, it is possible to roughly determine the written characters, pre-printing, and background. Specifically, for the pixels determined to be ruled lines in step 710, the determination criteria for the pre-print colors are relaxed on the premise that they are pre-print colors, that is, ruled line colors. For pixels in the character line area, the criteria for character color are relaxed. This process can prevent blurring of characters and ruled lines. Further, when the difference between the pre-print color and the background color and the character color and the background color are small, the separation standard can be tightened. This process can prevent blurring even when preprinted characters or ruled lines and the background have similar colors, and can prevent characters on a dark background from being crushed.

  As described above, by switching a plurality of separation criteria by the processing of FIG. 6, it is possible to generate a highly accurate binary image according to the purpose.

(Second Embodiment)
The second embodiment will be described with reference to FIG. In the second embodiment, the order and processing contents of the layout analysis (step 600) and background color detection (step 610) in FIG. 6 in the first embodiment are different.

  First, in step 610, the entire form image is divided into partial areas to perform background color detection. Next, in step 1100, layout analysis is performed using the detected background color information. Step 620 and subsequent steps are the same as those in FIG.

  In the second embodiment, the binary image generation process of the layout analysis process in step 1100 is different from step 700 in the first embodiment. Specifically, the binarization threshold value is changed for each pixel according to the background color. That is, when the background color is light (light), the binarization threshold is increased, and when the background color is dark (dark), the binarization threshold is decreased. The purpose of this processing is to improve the accuracy of layout analysis in a form in which different background colors are mixed as shown in FIG. In the first embodiment, it has been described that the threshold setting of the binarization process in step 700 of FIG. 7 is easier than the binarization threshold setting of FIG. However, depending on the difference in density between the background color and the pre-print color, it may not be possible to extract ruled lines or pre-print characters of all regions.

  FIG. 12 shows the result of detecting the background color of the input image of FIG. The background color of the area 220 in FIG. 2 is detected as 1210 in FIG. 12, and the area 230 is detected as 1220 in FIG. By blackening pixels of a color darker than the background color, a binary image for layout analysis that solves the above problem can be generated.

(Third embodiment)
A third embodiment will be described with reference to FIG. In the third embodiment, after the layout analysis (step 600) in FIG. 6 in the first embodiment, the region is divided in step 1300, and the processing after the background color detection in step 610 is performed for each region (step 1310). The purpose of this process is to improve the accuracy of binarization in a form in which the background color or the pre-print color has different hues or shades. Since it is difficult to set a separation reference in a situation where a plurality of pre-print colors and background colors are mixed, the separation reference is obtained after dividing the region into regions where the number of colors can be limited. As a specific example of area limitation, there is a method of dividing an area in frame units detected by frame extraction in step 720. This is based on the assumption that the background color is the same in the frame and the number of preprinted characters and ruled lines is small. In this method, all areas outside the frame are treated as the same area. As another example of the area division, there is a technique of dividing into rectangular areas of M × N pixels.

(Fourth embodiment)
The fourth embodiment will be described with reference to FIG. In the fourth embodiment, the region division in step 1300 is performed after step 1100 in the second embodiment. The processing after the preprint color detection in step 620 is performed for each region (step 1400). The purpose of this processing is to improve the accuracy of binarization in a form in which the hue and shade of the background color and pre-printed color are partially different in addition to improving the accuracy of layout analysis. Also in this embodiment, the background color may be obtained again for each region as in FIG.

(Fifth embodiment)
The fifth embodiment will be described with reference to FIG. In the fifth embodiment, after step 1100 of the second embodiment, pre-print color detection in step 620 and character color detection in step 630 are performed, and region division is performed in step 1500 using color information and layout information. The processing after the generation of the separation reference in step 640 is performed for each region (step 1510). As a specific example of area limitation, there is a method in which all frames having the same combination of background color, preprint color, and character color are set to the same area. Even outside the frame, the region is divided according to the combination of the above colors. In the present embodiment, color detection may be performed again for each region as in FIG.

(Sixth embodiment)
The sixth embodiment will be described with reference to FIGS. 16 to 23. The sixth embodiment is intended only to whiten the background color pixels and blacken the character color and pre-print color pixels. In the first to fifth embodiments, the binary images shown in FIGS. 10 (b) to 10 (d) can be generated for the input shown in FIG. 10 (a). Only the generation of the binary image of (b) is performed. In the present embodiment, a plurality of criteria for separating the background color and other colors are calculated, and binarization is executed by switching these criteria.

  The configuration of the image processing apparatus according to the sixth embodiment is the same as that of the first embodiment shown in FIG. However, the first embodiment differs from the first embodiment in that the function of the image processing unit 152 has the processing flow of FIG. Hereinafter, details of functions of the image processing unit 152 of the sixth embodiment will be described.

  An example of the binarization processing flow in the sixth embodiment is shown in FIG. First, the background color is detected for each of the divided partial areas using the same process as in step 610 of FIG. Next, in step 1640, which is a process corresponding to step 640 in FIG. 6, a reference for separating the background color pixel from the other pixels is calculated. Finally, in step 1680, which is a process corresponding to step 680 in FIG. 6, the separation criterion is switched for each pixel and binarized. Step 1640 will be described later with reference to FIG. 17, and step 1680 will be described later with reference to FIGS.

  Step 1640 in FIG. 16 is a process for calculating a reference for separating the background color from the other colors (character color and pre-print color). The separation criterion is calculated for all pixels to be binarized. For the calculation of the separation standard, brightness, brightness, saturation, hue, and the like are used. In this embodiment, an example using the luminance of the background color is shown. First, the reference values α and β are set in advance so that if the luminance of a pixel is less than or equal to α, blackening is always performed in the binarization process, and if the luminance is greater than or equal to β, whitening is always performed. Next, weights p and q (0.0 to 1.0) are set, and three separation criteria (T1 to T3) based on luminance are calculated using Equations 6 to 8. A specific description of T1 to T3 is as follows. T1 is a constant α that is a value forcibly blackened.

. T2 and T3 are values that change according to the luminance of the background color, and are used to determine a pre-print color or noise that is close to the background color. T3 is the smaller value of β and the value obtained by multiplying the luminance (Yb) of the background color at the position of the processing target pixel by the weight p and β.

. T2 is a value that internally divides T1 and T3 by q: (1-q).

. As an example, FIG. 17 shows T1 to T3 when α = 160, β = 220, p = 0.9, and q = 0.5. Note that the weights p and q may be constants or functions. Furthermore, the value of T3 may be a constant such as β that does not depend on the background color.

  Step 1680 in FIG. 16 includes three steps as shown in FIG.

  FIG. 19 shows a processing flow of pixel classification in step 1800 of FIG. The process of FIG. 19 is executed for all pixels in the binarization target area of the input image (steps 1900 and 1905). If it is determined in step 1910 that the luminance value of the pixel is less than T1, the pixel to be blackened is classified as C1 (blackening confirmed) in step 1920. If it is determined in step 1930 that the luminance value is equal to or greater than T1 and less than T2, it is classified as C2 (blackening candidate 1) in step 1940. If it is determined in step 1950 that the luminance value is greater than or equal to T2 and less than T3, it is determined whether or not another separation criterion is satisfied in step 1960. If the criterion is satisfied, C3 (blackening is determined in step 1970. Classify as candidate 2). If none of the criteria is satisfied, the pixel is classified as C4 (whitening confirmed) in step 1980 as a pixel to be whitened. As a specific example of the determination in step 1960, the difference between the lightness (Vb) of the background color and the lightness (Vi) of the pixel can be used.

Standard: Vb−Vi> γ
Where γ is a constant. This is because generally the background color is often lighter than the preprint color. Other conditions include the difference between the hue (Hb) of the background color and the hue (Hi) of the pixel, the difference between the saturation (Sb) of the background color and the saturation (Si) of the pixel, the lightness (Vi) of the pixel, An upper limit value of saturation (Si) can also be used. These conditions may be singular or a plurality of combinations.

  In step 1810 in FIG. 18, binarization is performed based on the classification result in step 1800. As an example of binarization, pixels classified as C1 to C3 are blackened, and pixels classified as C4 are whitened. As another example, pixels classified into C1 and C2 are blackened, and pixels classified into C3 and C4 are whitened. Note that steps 1800 and 1810 may be executed simultaneously by performing binarization (determination of blackening and whitening) immediately after classification for each pixel. In step 1820 of FIG. 18, based on the ratio of the pixels of each class, the result binarized in step 1810 is evaluated according to another criterion and re-binarized.

  The re-binarization process flow will be described with reference to FIG. In the example of FIG. 20, the description is based on the assumption that the pixels C1 to C3 are blackened in step 1810. In the processing of FIG. 20, the processing target area of the image is divided into rectangular areas of Mb × Nb pixels, and is executed for pixels in all areas (steps 2000 and 2005). Here, the size of the divided rectangular area may be the same as or different from the size at the time of background detection. In step 2010, the number of pixels classified into C1 to C3 in the region is counted. The counted number is defined as follows.

n1: Number of pixels classified as C1 (blackening decision) n2: Number of pixels classified as C2 (blackening candidate 1) n3: Number of pixels classified as C3 (blackening candidate 2) n = n1 + n2 + n3 (Total number of pixels of C1, C2, and C3)
n0 = Mb · Nb (total number of pixels in the region)
In step 2020, the following ratio is calculated in the region. These ratios are used in condition 1 in step 2030 and condition 2 in step 2050.

R = n / n0
R1 = n1 / n
R12 = (n1 + n2) / n
An example of condition 1 of step 2030 is shown below.
Condition 1: (R> t1) AND (R1> t2)
However, t1 and t2 indicate that the ratio of C1 (determined blackening) is high in the region under the constant condition 1 determined in advance. Therefore, the brightness of the character pixels in the region is sufficiently low, and the pixels with high brightness of C2 and C3 can be regarded as noise. Therefore, when the condition 1 is satisfied, in step 2040, only the pixels classified as C1 are blackened, and re-binarization processing is performed to whiten the pixels C2 and C3.

  The condition 1 will be specifically described with reference to FIG. FIG. 21 is an example in which the characters “ta” are written in an 8 × 8 (Mb = Nb = 8) area. FIG. 21A shows an example in which a pixel of a character classified as C1 such as a pixel 2110 exists in a background color region with a high luminance classified as C3 such as a pixel 2100. In addition, as a pixel classified as C2, there is a background noise (2120) and a background (2130) having a luminance value lower than that of other background colors. The reason why the pixel 2130 has a lower luminance value than the other background colors is because the color difference between the adjacent pixels is large. In such a case, the luminance value may be different from the actual value due to the influence of adjacent pixels due to the quantization error of the image sensor. FIG. 21C shows a result obtained by binarizing FIG. 21A using condition 1 and blackening only C1. Furthermore, as shown in FIG. 21B, the re-binarization result in the case where a character (C1) with low brightness is described in the background area (C4) with high brightness is also shown in FIG. 21 (c).

An example of condition 2 of step 2050 is shown below.
Condition 2: ((R> t3) AND (R2> t4))
OR ((R <t5) AND (n1 = 0))
However, from t3 to t5, the first line of the predetermined constant condition 2 means that the ratio of C1 and C2 in the region is high. The second row means that there is no C1 pixel with low luminance and the ratio between C2 and C3 is low. Therefore, the C2 pixel is a pre-printed character or ruled line having a color close to the background color and can be regarded as a blackening target, and a pixel having a high luminance of C3 is noise and can be regarded as a whitening target. Therefore, when the condition 2 is satisfied, in step 2060, re-binarization processing is performed in which the pixels classified into C1 and C2 are blackened and the pixel of C3 is whitened.

  Condition 2 will be specifically described with reference to FIGS. 22 and 23. FIG. 22 shows an example in which the first line of condition 2 is applied. FIG. 22A shows an example in which a pixel of a character classified as C2 such as a pixel 2210 exists in an area of a background color with high luminance classified as C3 such as a pixel 2200. The pixel C2 can be regarded as a preprinted character close to the background color. In FIG. 22B, in the background color area classified as C4 such as the pixel 2220, the character pixel classified as C2 such as the pixel 2230, and the background (2240) classified as C3 and having a low luminance value. It is an example that exists. FIG. 21C shows the result of binarizing FIG. 22A and FIG. 22B using the first line of condition 2. FIG. 23 shows an example in which the second line of condition 2 is applied. FIG. 23A shows a part of the pre-printed character classified as C2 such as the pixel 2310 and the noise classified as C3 (2320) in the background color area classified as C4 such as the pixel 2300. This is an example. FIG. 23B shows a rebinarization result in which C1 and C2 are blackened using the second line of condition 2. FIG. 23C shows an example in which noise classified as C3 exists in the background color area classified as C4 such as the pixel 2330. FIG. 23D shows the result of binarizing FIG. 23C.

  In an area where neither Condition 1 nor Condition 2 is satisfied, the pixels of C1, C2, and C3 remain blackened. Note that Condition 1 and Condition 2 are examples, and other conditions such as the average and dispersion of luminance, hue, saturation, brightness, and individual pixel values in the region can be used. Further, the number of conditions may be changed.

  As described above, in the sixth embodiment, binarization that whitens only the background color can be realized by using the difference from the background color and the luminance distribution in the small area.

  In FIG. 18, the binarization process may be terminated in step 1810. Further, instead of step 640 and step 680 of FIGS. 6, 11, 13, 14, and 15, step 1640 and step 1680 of this embodiment may be used.

(Seventh embodiment)
In the seventh embodiment, an example of an image processing apparatus that can perform not only binarization but also character recognition will be described. FIG. 24 is an example of a configuration diagram illustrating an image processing apparatus 2400 having the functions of the image processing apparatus 100 according to any one of the first to sixth embodiments.

  As shown in FIG. 24, the image processing apparatus 2400 has the same configuration as that of the image processing apparatus 100 shown in FIG. 1 except that a character recognition unit 2411 is added to the control unit 2410. To do. The character recognition unit 2411 performs character recognition processing from an image generated by binarization.

  For example, the character recognition unit 2411 performs binarization processing from the image of FIG. 2 and recognizes “amount”, “1”, “2”, “3”, “4”, “5”, and the like as characters. . The process for recognizing individual characters can use an existing method of pattern recognition.

(Eighth embodiment)
In the eighth embodiment, a store counter terminal device installed at a store counter of a financial institution using the image processing apparatus 100 according to any one of the first to sixth embodiments will be described. In financial institutions, etc., images of transaction forms entered by customers are captured by scanners connected to terminal devices installed at sales offices, and the resulting image data is stored and character recognition is performed on the image data. It is. As shown in FIG. 25, the sales office window terminal device 2500 includes an operation unit 110 (similar to FIG. 1), a display unit 120 (similar to FIG. 1), and an imaging unit 130 such as a stand-type scanner (similar to FIG. 1). ), A cash deposit / withdrawal unit 2510 that handles cash for exchange with a user, a passbook printer unit 2520 that prints transaction contents on a passbook, a printer unit 2530 that prints on receipts, a terminal storage unit 2540, and the like A terminal control unit 2550 for controlling. The terminal control unit 2550 corresponds to the control unit 150 in FIG. 1, and the terminal storage unit 2540 corresponds to the storage unit 140 in FIG.

  Hereinafter, a business flow using the sales office window terminal device 2500 will be described. The sales office window terminal device 2500 is a display unit 120 in which an operator who receives a transaction form for each transaction item from a user inputs a screen number for calling a transaction data input screen corresponding to the form from the operation unit 110. The target transaction is executed by inputting the entry data on the form such as the transaction amount into each input field of the transaction screen displayed on the terminal screen.

  Further, the terminal control unit 2550 collects an image of the transaction form collected by the scanner which is the imaging unit 130, reads the description content by an OCR (Optical Character Recognition) function based on the image, and displays a confirmation screen of the transaction content Displayed on the unit 120. In this case, the transaction details confirmation screen displays the contents of each data item read from the form by OCR, for example, the store number, account number, deposit amount, etc. Are transmitted to the management apparatus as recording data. In order to facilitate the confirmation of the recognition result, the input image of FIG. 10 (a) and one or more binary images of FIGS. 10 (b) to 10 (d) are displayed on the confirmation screen. May be.

  Thus, according to the branch office terminal device 2500, the OCR process can be executed from the image collected by the imaging unit 130.

DESCRIPTION OF SYMBOLS 100 Image processing apparatus 110 Operation part 120 Display part 130 Imaging part 140 Storage part 150 Control part 151 Imaging process part 152 Image processing part 2400 Image processing apparatus 2410 Control part 2411 Character recognition part 2500 Office counter terminal device 2510 Cash deposit / withdrawal part 2520 Passbook printer unit 2530 Printer unit 2540 Terminal storage unit 2550 Terminal control unit

Claims (3)

An image processing apparatus that generates a binary image from a color image obtained by imaging with a scanner,
A storage unit for storing color separation criteria for generating the binary image;
Obtaining color information of a plurality of colors of the color image by color clustering;
Two or more color separation criteria are calculated based on the obtained color information of the plurality of colors,
Based on the obtained color information of the plurality of colors and the color separation criterion, the color separation criterion is switched to perform binarization processing of the color image,
An image processing unit for storing the binarized pixel values in the storage unit;
With
The image processing unit
Dividing the color image into partial areas;
Obtain the background color of each area by color clustering,
Calculating a plurality of color separation criteria for the background color and other colors;
Perform binarization processing by switching the color separation standard,
Calculating a plurality of separation criteria based on the difference between the color of each pixel and the background color;
Classifying pixels into a plurality of classes using the plurality of separation criteria;
Dividing the region into partial regions, counting the number of pixels belonging to each class in each partial region, obtaining an existence ratio, and determining a class to be blackened based on the existence ratio An image processing apparatus.   An image processing method for generating a binary image from a color image obtained by imaging with a scanner,
  Using a storage unit that stores color separation criteria for generating the binary image,
  Obtaining color information of a plurality of colors of the color image by color clustering;
  Two or more color separation criteria are calculated based on the obtained color information of the plurality of colors,
  Based on the obtained color information of the plurality of colors and the color separation criterion, the color separation criterion is switched to perform binarization processing of the color image,
  An image processing method for storing the binarized pixel values in the storage unit, further comprising:
  Dividing the color image into partial areas;
  Obtain the background color of each area by color clustering,
  Calculating a plurality of color separation criteria for the background color and other colors;
  Perform binarization processing by switching the color separation standard,
  Calculating a plurality of separation criteria based on the difference between the color of each pixel and the background color;
  Classifying pixels into a plurality of classes using the plurality of separation criteria;
  Dividing the region into partial regions, counting the number of pixels belonging to each class in each partial region, obtaining an existence ratio, and determining a class to be blackened based on the existence ratio Image processing method. A program for generating a binary image from a color image obtained by imaging with a scanner,
Obtaining color information of a plurality of colors of a color image input by color clustering;
Calculating two or more color separation criteria based on the obtained color information of the plurality of colors;
Performing binarization processing of the color image by switching the color separation reference based on the obtained color information of the plurality of colors and the color separation reference;
Storing the binarized pixel values in a storage unit;
Is a program characterized by causing a computer to execute,
Dividing the color image into partial regions;
Obtaining a background color for each region by color clustering;
Calculating a plurality of color separation criteria for the background color and other colors;
Performing binarization processing by switching the color separation reference;
Calculating a plurality of separation criteria based on the difference between the color of each pixel and the background color;
Classifying pixels into a plurality of classes using the plurality of separation criteria;
Dividing the region into partial regions, counting the number of pixels belonging to each class in each partial region, obtaining an existence ratio, and determining a class to be blackened based on the existence ratio; A program characterized by being executed by a computer.