JP2005092543A - Form reader - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a form reader for reading a form sealed in a certain position, where the seal overlaps characters written in a color different from the color of the seal, in such a way that the seal is dropped while a character string written in the same color as the seal is not dropped out so that the written characters having the different color and overlapping the seal and the written characters having the same color and not overlapping the seal are read accurately. <P>SOLUTION: A sealed area is detected and the seal is eliminated from the color image of the sealed area to create a binary image. The binary image from which the seal has been eliminated is pasted and combined with an inputted monochrome binary image. Characters are recognized on the basis of the resulting composite binary image. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a form reading apparatus that generates an image from which a seal is removed from a form with a red seal stamp and reads characters.

  When a form image is collected by a scanner and character recognition is performed, an imprint in the form may overlap with characters and interfere with character recognition. Therefore, it is necessary to remove the imprint in the form that becomes an obstacle to character recognition. Conventionally, in Japanese Patent Laid-Open No. 5-28313 (Patent Document 1), as a method for removing an imprint, a red filter and a blue filter are provided together, and the red imprint is dropped out by the red filter. Yes. Conventionally, JP-A-8-212294 (Patent Document 2) describes that an imprint reading mode and a character reading mode are provided, and imprint color image data is generated using only the color component of the imprint as effective dots. Yes. Conventionally, in Japanese Patent Application Laid-Open No. 9-81666 (Patent Document 3) and Japanese Patent Application Laid-Open No. 2002-183661 (Patent Document 4), when the color of the seal is different from the color of the written character, the color portion of the seal is separated by color separation. It is described that it is determined whether or not a seal is present based on the position where the strikethrough is written. Conventionally, Japanese Patent Laid-Open No. 2000-251072 (Patent Document 5) describes that a pre-printed image is removed from a form image including pre-printing of the same color as that of a stamp and a stamp is extracted. Conventionally, Japanese Patent Laid-Open No. 2002-269547 (Patent Document 6) describes that a black and white image is used to distinguish a square mark from a stamped column frame and extract a seal image. Conventionally, in Japanese Patent Application Laid-Open No. 8-221507 (Patent Document 7), an imprinted part stamped at the end of a sentence is verified by comparing it with a character recognition result and a word that may appear at the end of the sentence. It is described that an imprinted part included is removed.

Japanese Patent Laid-Open No. 5-28313

JP-A-8-212294 JP-A-9-81666 JP 2002-183661 A Japanese Patent Laid-Open No. 2000-255102 JP 2002-269547 A JP-A-8-221507

  However, in the conventional method in which the red imprint is dropped out by the red filter and the imprint color image data is generated using only the color component of the imprint as effective dots, the red character, that is, the character of the same color as the imprint Is erased, there is a problem that it is impossible to read the same characters as the seal. Further, in the conventional method for determining whether or not a seal exists on the basis of a position where a strikethrough is described, there is a problem that extraction of a seal imprinted at an arbitrary position without a strikethrough is not considered. . In addition, the conventional method of extracting imprints by distinguishing between square marks and stamped field frames is based on the premise that the stamps are stamped in the vicinity of the stamped field frames, and the stamps imprinted at arbitrary positions on the form are extracted. There is a problem that is not considered. In addition, in the conventional method of extracting a stamp image by removing a pre-printed image from a form image, it is necessary to prepare a pre-printed image in advance. There is a problem that providing images is complicated. In addition, the conventional method of verifying an imprinted part stamped at the end of a sentence by comparing the character recognition result with a word that may appear at the end of the sentence, and removing the imprinted part included in the character recognition result, The overlap is not considered, and if the seal and the character overlap, there is a problem that it is difficult to remove the seal portion compared to a word that may appear at the end of the sentence.

  The present invention has been made in view of the above-mentioned problems, and its purpose is to read a form in which a seal stamp is stamped at an arbitrary position and a form in which an input character different from an imprint color and an imprint are overlapped. , Without dropping out the character string written in the same color as the imprint color, drop out the imprint, and the form reading device that accurately reads the different colored input characters that overlap the imprint and the same color that does not overlap the imprint Is to provide.

  The present invention detects an imprint area where a seal is stamped, generates a binary image in which the imprint is erased from the color image of the imprint area, and pastes the binary image from which the imprint has been erased on the input black and white binary image. The character is recognized based on the combined binary image.

  Further, in order to detect an imprinted region where the seal is imprinted, the present invention extracts pixels of the same system as the seal color from the color image, extracts a block of the pixels as a connected component, and defines a circumscribed rectangle of the connected component. The size of the circumscribed rectangle or the density of red pixels in the circumscribed rectangle is detected, and the circumscribed rectangle is extracted as an imprint area based on the size or density.

  Further, the present invention generates a binary image in which the imprint is erased from the color image in the imprint region, so that one primary color component is selected from the three primary color components of the color image, and each pixel is regarded as the density value. When the binarization result of the pixel of interest becomes black, the pixel of interest is color-identified, and the pixel of interest determined to be the primary color system selected by the color identification result is inverted to white. I have to.

  The present invention also provides means for displaying, for each reading field, a black and white binary image obtained by pasting and synthesizing a partial binary image from which an imprint has been deleted to the input black and white binary image, and reading the input color image for each reading field. And a means for switching the composite black-and-white binary image or color image for each reading field for each field, detecting the presence or absence of overlap between the reading field and the imprint area, and the reading field as an imprint area. The input color image is displayed in the overlapping field.

  According to the present invention, it is possible to detect an imprint area from a form including characters of the same color as the seal color. Therefore, it is possible to read a form having an entry character of the same color as the seal and to read various forms. There is an effect.

  In addition, according to the present invention, a binary partial image in which an imprint area is extracted and the imprint of the area is removed can be generated and pasted and synthesized on the original black and white binary image. There is an effect that a character overlapped with an imprint can be accurately read in a binary image.

  In addition, according to the present invention, since the imprint region is limited and detected by the size and the density of red pixels, even if there is a partial red border or red entry character, it is not removed, This has the effect of removing only the imprint.

  Further, according to the present invention, when the reading result is corrected via the screen, the color image to be displayed and the binary image can be switched depending on whether or not the imprint is overlapped. There is an effect that the read result can be corrected accurately even if the form has a certain form.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a configuration diagram of a form reading apparatus according to an embodiment of the present invention. The color image input unit 100 collects a surface image of a form as a two-sided image of a color image 110 and a monochrome binary image 111. Here, the black-and-white binary image 111 is generated by binarization processing from an image in which the minimum value of the three primary color components is a grayscale image in each pixel. Since the minimum values of the three primary color components are binarized as a grayscale image, all colored pixels and black dark pixels are black in the binarized image. Accordingly, the imprinted character, the character of the same color as the imprint, the character of different color, and the ruled line become a black binary image.

  The image processing unit 101 includes an imprint region detection unit 103 that detects the region coordinates of the imprint, and a region color image binarization unit 104 that generates a binarized image by cutting out a color image in the imprint region and dropping out the imprint color. The image pasting / combining unit 102 pastes the binarized result of the color image in the region to the black and white binary image 111. In the collected color image 110, the imprint area detection unit 103 extracts the coordinates of the imprint area with respect to the imprint imprinted at an arbitrary position. Then, based on the area coordinates, the area color image binarization unit 104 binarizes the color image. Here, for each pixel, color binarization is performed in which a pixel having an imprint color is dropped out to be white and a pixel having another color is black. The image pasting / combining unit 102 replaces and pastes the imprint area extracted inside the black and white binary image 111 with the partial image which is the result of binarizing the area color image, thereby pasting the new black and white binary image 112. Generate. The synthesized black and white binary image 112 is a binary image in which the imprint area is a dropout of the imprint color, and the other portion is a binary image binarized with the minimum value of the three primary color components as a grayscale image. It becomes.

  Characters of the synthesized black and white binary image 112 output from the image processing unit 101 are read by the character recognition unit 105 and converted into character codes. The character correction unit 106 corrects the character recognition result read by the character recognition unit 105 by displaying a form image on the screen. In the seal imprint area, the characters that are different in color from the seal imprint are black, and the seal imprint is dropped and becomes white. Therefore, there is an effect that the character recognition unit 105 can accurately read the characters that overlap with the seal imprint as black. In addition, the same or different colored characters that do not overlap with the seal are black, and the character recognition unit 105 can read the characters accurately.

  A form reading apparatus that performs these processes includes a color image input unit, an image processing unit, a character recognition unit, and a character correction unit. The color image input unit has a scanner of the optical reading device as an input unit. The image processing unit performs processing of detecting an imprint region, binarizing a region color image, and pasting and combining partial images based on image data read by the input unit. In addition, the character recognition unit reads the character. These image processing unit and character recognition unit are configured by a small computer program. This program is stored in a storage device such as a hard disk of a form reading device or a CD-ROM. In addition, the character correction unit is composed of a display device such as a CRT that displays the result.

  FIG. 2 is an example of a color form image targeted by the present invention. The form image 200 is an example of a color image 110 and includes character strings 201, 203, 204, 205, 206, an imprint 207, and a frame line 202. The imprint color is red as indicated by the imprint 207, and the text color is red and black. The entered character string 201 that does not overlap the seal is red. In addition, the color of the input character strings 204 and 206 overlapping the seal impression is black. The rightmost part of the amount and the rightmost part of the payer overlap the imprint.

  FIG. 3 is an example of a black and white image that is the subject of the present invention, and is an image of one of the two images simultaneously collected by the color image input unit 100. The black-and-white image 300 is an example of a black-and-white binary image 111 generated by binarization processing from an image in which the minimum value of the three primary color components is a grayscale image at each pixel. The character strings 301, 303, 304, 305, 306 and the frame line 302 are black. Also, the imprint 307 is also black. The entered character strings 304 and 306 overlap with the seal imprint 307, and if the black and white image 300 is directly read by the character recognition unit 105 as in the conventional method, it is difficult to accurately read the characters because the seal imprint 307 is an obstacle. It is. Therefore, an imprint area is detected using color information, and color binarization processing is performed on the detected imprint area.

  FIG. 4 is a diagram for explaining the detection result of the seal impression area. A result image obtained by extracting an imprint region with respect to the color image 200 is indicated by 400. An imprint area detected by the imprint area detection unit 103 is indicated by 407. The imprint area is output in rectangular vertex coordinates.

  FIG. 5 is a binary image obtained as a result of binarizing the color image of the seal impression area. The area color image binarization unit 104 cuts out the color image inside the imprint area 407 and binarizes the cut out partial color image by dropping out red color to obtain a result image 500. The imprinted portion is erased to be white, and the entered character string portion and the frame line are black.

  FIG. 6 shows an example of the black and white binary image 112 synthesized by the image processing unit 101, and the partial image that is the result of binarization of the above-described area color image in the imprint area extracted inside the black and white binary image 111. Is a new black-and-white binary image pasted. A new black and white binary image pasted with a partial image as a result of binarizing the area color image is indicated by 600. The red entry character string 601 is black. Also, the black and white character strings 604 and 606 are erased, and all the characters are black. In this way, the character string of the same color that does not overlap with the seal imprint is black without being erased, and the character string with a different color that overlaps the seal imprint is black without being erased, so that the character string with the seal imprint can be accurately read. There is an effect.

  FIG. 7 is a diagram for explaining the process of the seal impression area detection unit 103. The input is the three primary color components of the color image, and the output is the coordinates of the impression area. First, in step 700, the imprint region detection process is started, and in step 701, the three primary color components are input. Next, in step 702, red pixels are extracted. Here, by comparing the sizes of the three primary color components of each pixel, a pixel having a large red component is defined as a red pixel. In step 703, the number of red pixels is counted on the entire surface of the form image, and in step 704, it is determined whether the number of red pixels is larger than a predetermined value. If the number of red pixels is larger than the predetermined value, the main imprint area detection process is rejected in step 705 and the process ends in step 706. On the other hand, if the number of red pixels is smaller than the predetermined value, in step 707, a black image block is merged with a binary image in which the red pixels are black and the other pixels are white, and extracted as a connected component. The circumscribed rectangle of the merged connected component becomes a candidate for the imprint region. Next, in step 708, the size of the circumscribed rectangle of the connected component is tested. Here, based on a predetermined size range of the seal impression, it is tested whether the size of the circumscribed rectangle of the connected component satisfies a predetermined size condition. By this step, it is determined that a horizontally long region such as a red entry character string is not an imprinted region. Further, it is determined that a small area such as a red entry character is not an imprint area. In this step, even if an entry character has the same color as the seal imprint, it is not considered as a candidate for the seal imprint area, so that the entry character of the same color that does not overlap the seal impression is not erased. Further, in step 709, the pixel density in the circumscribed rectangle of the connected component is measured, and it is assumed that the region with a low pixel density is not a candidate for the imprint region. On the other hand, an area having a high pixel density in the rectangle is set as a seal impression area candidate. In step 710, the coordinates of a plurality or a single imprint region candidate are output, and in step 711, the imprint region detection process is terminated. In this way, in the imprint area detection process, the candidate area is obtained, and the candidate area is tested based on the size of the candidate area and the pixel density in the candidate area, so that the imprint area is distinguished from the character area of the same color. This has the effect of detecting only the seal impression area.

  FIG. 8 is a diagram for explaining the processing steps of the area color image binarization unit 104. In step 800, processing for binarizing the color image in the seal impression area is started. First, in step 801, the coordinates of the seal impression area candidate are input. The coordinates of the region may be rectangular vertex coordinates. In step 802, the color image in the region is cut out as a partial image. In step 803, a red component image is selected from the three primary color components of the color partial image, and is regarded as a grayscale image to be binarized. In step 804, scanning of the image in the vertical direction is repeated, and in step 805, scanning of the image in the horizontal direction is repeated. In step 806, the grayscale image value of the target pixel is binarized. Here, assuming that black has a dark gray value and white has a light gray value, the pixel of interest is binarized to white or black. In step 807, it is determined whether the pixel of interest is black or white. If the pixel of interest is black, color identification of the pixel of interest is performed in step 808. Here, based on the three primary color components of the pixel of interest, the value of the red component is compared with the values of the other two primary color components, i.e., the blue component and the green component. If larger, the target pixel is determined to be red. In step 809, if the target pixel is red, in step 810, the target pixel is whitened, that is, replaced with white. Since the output of this process is in the form of a bit-filled black and white binary image, bit packing is performed on the 1-byte image data in units of 8 black and white pixels in step 811. In step 812, the region color image binarization processing is terminated. In this process, in step 803 and step 806, not only the red component of the three primary color components is selected and binarized as a grayscale image, but also if the target pixel is red in step 810, the target pixel is whitened. There is an effect that the red color of the form image can be more accurately dropped out. For this reason, the red stamp is dropped out to become white, and the entered characters and the frame line become black. The portion where the seal mark overlaps with the entered character or the frame line is not dropped out because it is not determined to be red from the comparison result of the values of the three primary color components, and becomes black as a result of the binarization process.

  FIG. 9 is a diagram for explaining the process of the image pasting / combining unit 102. Here, the black-and-white binary image 111 and the binary image of the impression image color image which is a partial image are pasted together to generate a combined black-and-white binary image 112. In step 900, the image pasting synthesis process is started. First, in step 901, the upper left vertex coordinates of the seal impression area are input. In step 902, the width and height of the seal impression area are input. In step 903, the image is scanned in the vertical direction, and in step 904, the process from steps 905 to 908 is repeated while the image is scanned in the horizontal direction. First, in step 905, the pixel position of the pasting destination is calculated. Next, in step 906, the pixel position of the pasting source is calculated. In step 907, the value of the target pixel of the pasting source is read, and in step 908, the value of the target pixel is embedded in the pasting destination pixel. The above processing is performed while moving the pixel of interest in the vertical and horizontal directions. In step 909, the image pasting synthesis process is terminated. In this way, since the binary image in which red is dropped out is pasted only on the partial image in the seal impression area, there is an effect that the red character string is not dropped out.

  FIG. 10 is a diagram for explaining the overlap between the reading field and the seal area. The form image 1000 is an example of the black and white binary image 111. In the form image 1000, reading fields for reading characters are 1001, 1002, and 1003. The imprint area is 1004. The reading field can be given as a format in the case of a standard form. Further, in the case of a semi-standard form with a slightly different layout, it is possible to extract a frame line and detect a reading field from the arrangement of the frame. The imprint area 1004 can be detected by the imprint area detection unit 103. In this embodiment, the reading fields 1002 and 1003 partially overlap the imprint area 1004. If this imprint and the reading field overlap, a human being corrects an error or rejection of the reading result, so even if the black and white image 1000 is displayed on the screen, a character with a character string written in the reading fields 1002 and 1003 However, it is difficult to read even visually. On the other hand, in the color image 110, even if the character string overlaps with the seal, it is easy for a human to read the character visually. In view of this, the character correction unit 106 switches the screen display of the reading field to two types, that is, a black and white image and a color image, so that even a character string with a seal imprint can be easily confirmed and corrected by human eyes. .

  FIG. 11 is a diagram for explaining the process of field image display processing by the character correction unit 106. In step 1100, field image display processing is started. In step 1101, the color image 110 is input, and in step 1102, the black and white image 110 or 112 is input. In step 1103, the position coordinates of the reading area are input as field information. In step 1104, the position coordinates of the seal impression area are input. The screen display of the field is sequentially performed in step 1105 until the field is exhausted. First, the field of interest coordinates are set in step 1106, and in step 1107, it is determined whether or not there is an overlap between the seal impression area and the field of interest. If the seal impression area and the target field overlap in step 1108, the target field is displayed as a color image in step 1107. On the other hand, if it is determined in step 1108 that the seal impression area and the target field do not overlap, in step 1110 the target field is displayed as a black and white image. With these screen displays, humans visually observe the screen and correct the reading result of the form. When the field is exhausted in step 1105, the correction of the reading result through the screen of the form ends in step 1111. . According to the character correction processing of the present invention, even in the case of a black-and-white image that is difficult for humans to read visually with a character string that overlaps with an imprint, the color image is switched and displayed, so that the character can be easily observed.

It is a block diagram of the form reading apparatus which is one Example of this invention. It is an example of the color form image made into the object of this invention. It is an example of a black and white image that is a subject of the present invention, and is an image of one of the two images that are simultaneously collected by the color image input unit 100. It is a figure explaining the detection result of an imprint area. It is a binary image as a result of binarizing the color image of the seal impression area. A new monochrome binary image which is an example of a synthesized monochrome binary image 112 and in which a partial image which is the result of binarization of the above-described area color image is pasted in an imprint area extracted within the monochrome binary image 111 It is. It is a figure explaining the process of the imprint area detection part. FIG. 6 is a diagram for explaining a processing process of an area color image binarization unit 104. It is a figure explaining the process of the image sticking synthetic | combination part. It is a figure explaining the overlap of a reading field and a seal stamp area. FIG. 10 is a diagram illustrating a processing process of a character correction unit 106.

Explanation of symbols

DESCRIPTION OF SYMBOLS 102 ... Image pasting composition part, 103 ... Imprint area detection part, 104 ... Area color image binarization part, 105 ... Character recognition part, 106 ... Character correction part, 110 ... Color image, 111 ... Monochrome binary image, 112 ... Black and white binary image after pasting, 200... Color document image, 201... Red input character, 204. Black input character, 206. Black input character, 207. Detected imprint region, 500 ... Color binarization result image of the imprint region, 600 ... Black and white binary form image after pasting and composition, 707 ... Merged connected component extraction processing step, 708 ... Size test of circumscribed rectangle of connected component Processing step 709 ... Pixel density test in circumscribed rectangle of connected component, 803 ... Selection of red component image, 808 ... Color identification of target pixel, 1002 ... Reading field, 004 ... imprint area, 1107 ... overlap determination of the field of interest with the imprint area.

Claims (8)

In a form reading apparatus having an image input unit for inputting image data, a recognition unit for recognizing characters from the input image data, and an output unit for outputting the recognition result,
The recognition unit generates a color image and a black and white binary image from the input image data, detects an imprint area where a seal is imprinted from the color image,
Generate a binary image in which the imprint is erased from the color image of the imprint area,
Pasting and synthesizing the binary image with the imprint deleted on the black and white binary image,
A form reading apparatus for recognizing characters based on the synthesized binary image.   2. The form reading apparatus according to claim 1, wherein the detection of the imprinted region is performed by extracting pixels of the same system as the seal color from the color image, extracting a block of the pixels as a connected component, and circumscribing the connected component. A form reading apparatus characterized in that a rectangle is obtained, a size of a circumscribed rectangle or a density of red pixels in the circumscribed rectangle is detected, and the circumscribed rectangle is extracted as an imprint area based on the size or density.   2. The form reading apparatus according to claim 1, wherein the binary image is generated by selecting one primary color component from the three primary color components of the color image, regarding the primary color component as a density value, and treating each pixel as a binary binary image. When the binarization result of the target pixel becomes black, the target pixel is color-identified, and the target pixel determined to be the selected primary color system is white-inverted. A form reader. In an image data correction apparatus having an image input unit for inputting image data, a correction processing unit for correcting an imprint region from the input image data, and a display unit for displaying the correction processing result,
The display unit displays a black-and-white binary image obtained by pasting and combining a partial binary image from which an imprint has been deleted on an input black-and-white binary image or an input color image by switching for each field.
The correction processing unit detects whether or not the reading field overlaps with the imprint region, and generates a color image in which the imprint region is corrected for the field where the reading field overlaps with the imprint region, and displays the color on the display unit. A form reading apparatus characterized by displaying an image. In the recognition unit of the form reading apparatus having an image input unit for inputting image data, a recognition unit for recognizing a form from the input image data, and an output unit for outputting the recognition result,
A first step of generating a color image and a black and white binary image from the input image data;
A second step of detecting an imprinted region where a seal is imprinted from the color image;
A third step of generating a binary image in which the imprint is erased from the color image of the imprint region;
A fourth step of pasting and synthesizing the binary image from which the imprint has been deleted to the monochrome binary image;
And a fifth step of recognizing a character based on the synthesized binary image. 6. The program according to claim 5, wherein the second step includes
Extracting a pixel of the same system as the seal color from the color image, extracting a block of the pixel as a connected component, obtaining a circumscribed rectangle of the connected component, a size of the circumscribed rectangle or a circumscribed rectangle A program comprising: detecting a density of red pixels; and extracting the circumscribed rectangle as an imprint area based on the size or density. 6. The program according to claim 5, wherein the third step includes
A step of selecting one primary color component from the three primary color components of the color image, a step of considering each primary color component as a density value and binarizing each pixel into black and white, and a binarization result of a target pixel is black The program further includes a step of color-identifying the pixel of interest, and a step of reversing the pixel of interest determined to be the selected primary color system by white. In an image data correction apparatus having an image input unit for inputting image data, a correction processing unit for correcting an imprint region from the input image data, and a display unit for displaying the correction processing result,
The correction processing unit detects the presence / absence of an overlap between the reading field and the imprint area for each reading field, and generates a color image in which the imprint area is corrected for the field where the reading field overlaps the imprint area. Step and
The display unit displays a black and white binary image obtained by pasting and combining a partial binary image from which an imprint has been deleted on the input black and white binary image or an input color image for each field. A program for executing the step of displaying the corrected color image for a field overlapping with an area.

Images