JP2014016921A - Form type discrimination apparatus and form type discrimination method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To precisely discriminate the type of forms bearing characters and graphics, a seal of approval, a barcode label, etc.SOLUTION: The form type determination method includes the steps of: extracting an edge by searching an image of a form in a y-direction to create a first edge image which is binarized based on the presence/absence of the edge (S3); creating a first compressed image by compressing the first edge image in an x-direction (S4); creating a first histogram in which black pixels present in the x-direction in the first compressed image are plotted at corresponding positions in the y-direction (S5); calculating a first matching degree by comparing the black pixels present in the x-direction in pre-registered various forms and a first standardized histogram plotted at corresponding positions in the y-direction (S6); and outputting a type discrimination result of the form on the basis of the first matching degree.

Description

  The present invention is capable of discriminating a form type such as a form having a small density difference between a background and a ruled line, a form on which characters / graphics are entered, a form on which a cashier stamp is pressed, or a form on which a barcode label is attached. The present invention relates to a type discrimination device and a form type discrimination method.

  2. Description of the Related Art Conventionally, there is known a form format identifying device that compares a form length distribution pattern obtained from a form image obtained by capturing an image with an image pickup apparatus with a pre-registered form length distribution pattern to obtain a candidate for a form type (patent) Reference 1).

  In this Patent Document 1, ruled lines are extracted from a form image to obtain a ruled line length distribution pattern and a ruled line interval, and the obtained ruled line length distribution pattern is compared with a standard ruled line length distribution pattern of various forms registered in advance. Thus, several types having a high degree of coincidence are candidates. Then, the ruled line interval of each candidate and the standard ruled line interval of each candidate registered in advance are divided into a plurality of divided areas and compared to obtain the expansion rate of each candidate form image. Based on the expansion rate, Normalize the ruled line spacing and ruled line position, calculate the ruled line spacing and ruled line position when there is no expansion and contraction, and calculate the candidate ruled line spacing and ruled line position and the standard ruled line spacing and standard ruled line position of each candidate registered in advance. It is disclosed that the type having the highest degree of coincidence is output as a discrimination result.

  Further, in Patent Document 1, ruled lines are extracted from a form image to obtain ruled line intervals, and the obtained ruled line intervals and standard ruled line intervals of various registered forms are divided into a plurality of divided areas and compared. Obtain the stretch rate of the candidate form image, normalize the ruled line spacing and ruled line position of each candidate based on the stretch rate, estimate the ruled line spacing and ruled line position when there is no stretch, and each candidate ruled line spacing and ruled line position And the standard ruled line interval and standard ruled line position of each candidate registered in advance, and the type with the highest degree of matching is output as a discrimination result.

JP 2000-123174 A

  However, in the above-described conventional form format identification device, for a form having a small density difference between the background and the ruled line, the ruled line is blurred on the image from which the ruled line is extracted, and the ruled line cannot be extracted well. There was a problem that the type could not be accurately identified.

  In addition, for images such as forms with letters / graphics, forms with stamps on and off, forms with barcode labels attached, etc., the characters, banknotes and barcode labels will be noisy. However, there is a problem that the ruled line cannot be extracted well and the discrimination accuracy of the form type is lowered.

  The present invention has been made to solve the above-described problems of the prior art, and is a form having a small density difference between the background and the ruled line, a form on which characters / graphics are entered, a form on which a cashier is pressed, or a barcode. It is an object of the present invention to provide a form type discriminating apparatus and a form type discriminating method capable of accurately discriminating a form type such as a form with a label attached thereto.

  In order to solve the above problems, the present invention is generated by a differential image generation unit that generates a differential image obtained by binarizing a form image of a form in the vertical direction or the horizontal direction, and the differential image generation unit. A compressed differential image generating unit that generates a compressed differential image obtained by compressing the differential image in the horizontal direction or the vertical direction, and a feature amount generating unit that generates a predetermined feature amount related to the compressed differential image generated by the compressed differential image generating unit; , A degree-of-match calculation unit that calculates a degree of match using a feature amount of a compressed differential image related to each form registered in advance and a feature amount generated by the feature amount generation unit, and the degree-of-match calculation unit Output means for outputting a result of discriminating the form type based on the degree of match.

  In the invention described above, the compressed differential image generation unit may set the pixel of the first predetermined length in the form as one pixel in the differential image generated by the differential image generation unit, and Only when the second predetermined number of pixels among the pixels having the first predetermined length are pixels that continuously form the edge portion, the pixel value of the pixel is set to the pixel value assigned to the edge portion. A minimum differential compression is performed in the horizontal direction or the vertical direction to generate a compressed differential image.

  Further, in the present invention according to the above invention, the feature amount generation unit may use, as the feature amount, a histogram obtained by cumulatively adding the number of pixels forming the edge portion included in the compressed differential image in the horizontal direction or the vertical direction. The degree-of-match calculation unit generates the degree of match using a standard histogram of a compressed differential image relating to each form registered in advance and the feature amount generated by the feature amount generation unit.

  Further, in the present invention according to the above invention, the feature amount generation unit may add a histogram obtained by cumulatively adding the number of pixels forming the edge portion included in the compressed differential image in the horizontal direction or the vertical direction to the standard histogram. It further comprises normalizing means for normalizing to the combined size.

  Further, according to the present invention, in the above invention, the matching degree calculation unit is a position where the peak position in the standard histogram matches the peak position in the histogram for each partition area set in the standard histogram of various forms. The histogram is corrected so as to match the calculated position, and the degree of match is calculated by comparing the corrected histogram with a standard histogram.

  Further, in the present invention according to the above invention, the output unit may determine a corresponding area on the form image corresponding to a print area set in relation to the form type having a high degree of matching based on the positional deviation amount. Then, a character in the corresponding area is extracted, and a form type that can be recognized as a print character set in relation to the form type having a high degree of coincidence is output.

  The present invention also provides a differential image that generates a first differential image obtained by differentiating a form image of a form in the vertical direction and binarized, and a second differential image obtained by differentiating the form form image in the horizontal direction and binarized. A first compressed differential image in which the first differential image generated by the differential image generation unit is compressed in the horizontal direction and a second compressed image in which the second differential image generated by the differential image generation unit is compressed in the vertical direction. A compressed differential image generating unit configured to generate a compressed differential image; a first feature amount relating to the first compressed differential image generated by the compressed differential image generating unit; and a second compressed differential image generated by the compressed differential image generating unit. A feature amount generating means for generating a second feature amount; a first feature amount and a second feature amount relating to each form registered in advance; and a first feature amount and a second feature amount generated by the feature amount generating means. And matching degree calculation means for calculating a degree of match you are, characterized in that an output means for outputting the type determination result of the form based on the matching degree calculated by the matching degree calculating unit.

  The present invention also provides a differential image generation step for generating a differential image obtained by differentiating a form image of a form in the vertical direction or the horizontal direction and binarizing the differential image generated by the differential image generation step in the horizontal direction or A compressed differential image generating step for generating a compressed differential image compressed in the vertical direction, a feature amount generating step for generating a predetermined feature amount related to the compressed differential image generated by the compressed differential image generating step, and each of the pre-registered A matching degree calculation step of calculating a matching degree using the feature amount of the compressed differential image relating to the form and the feature amount generated by the feature amount generation step; and a form based on the matching degree calculated by the matching degree calculation step And an output process for outputting the type discrimination result.

  According to the form type discriminating apparatus and form type discriminating method according to the present invention, a form having a small density difference between the background and the ruled line, a form on which characters / graphics are entered, a form on which a cashier stamp is pressed, or a bar code label is attached. A ruled line can be extracted with high accuracy from a form.

FIG. 1 is a block diagram illustrating a configuration of a form type discriminating apparatus according to the present embodiment. FIG. 2 is a flowchart (1) showing the processing of the form type discriminating apparatus according to the present embodiment. FIG. 3 is a flowchart (2) showing the processing of the form type discriminating apparatus according to the present embodiment. FIG. 4 is an exemplary diagram showing a form image G after tilt correction. FIG. 5 is an exemplary diagram showing the first edge image E1. FIG. 6 is an exemplary diagram showing the first compressed image F1. FIG. 7 is an exemplary diagram showing the first histogram H1. FIG. 8 is an exemplary diagram showing various forms Ca to Cd and first standard histograms Ha to Hd about them. FIG. 9 is an explanatory diagram showing the partitioned areas A1 to A7 registered in advance in relation to the first standard histogram Ha. FIG. 10 is an explanatory diagram showing positional deviation amounts y1 to y7 between the large peak of the first histogram H1 and the large peak of the first standard histogram Ha. FIG. 11 is an exemplary diagram showing the second edge image E2. FIG. 12 is an exemplary diagram showing the second compressed image F2. FIG. 13 is an exemplary diagram showing the second histogram H2. FIG. 14 is an exemplary diagram showing various forms Ca to Cd and second standard histograms Ja to Jd about them. FIG. 15 is an explanatory diagram showing partitioned areas B1 to B4 registered in advance in relation to the second standard histogram Ja. FIG. 16 is an explanatory diagram showing positional deviation amounts x1 to x4 between the large peak of the second histogram H2 and the large peak of the second standard histogram Ja. FIG. 17 is an explanatory diagram showing the horizontal printing area 15a and the vertical printing area 25a of the form Ca. FIG. 18 is an explanatory diagram showing a horizontal print area 15a ′ moved in the reverse direction by the positional deviation amount y1 and a vertical print area 25a ′ moved in the reverse direction by the positional deviation quantity x1. FIG. 19 is an explanatory diagram showing a region in which the moved horizontal print region 15a ′ and vertical print region 25a ′ are applied to the form image G.

  Embodiments of a form type discriminating apparatus and form type discriminating method according to the present invention will be described below with reference to the accompanying drawings. Note that the present invention is not limited thereby.

  First, the configuration of the form type discriminating apparatus 100 according to the present embodiment will be described. FIG. 1 is a block diagram showing the configuration of the form type discriminating apparatus 100. As shown in the figure, the form type discrimination device 100 includes a form image input unit 1, a form image inclination correction unit 2, a first edge image creation unit 11, a first feature detection unit 12, and a first feature. A collation unit 13, a second edge image creation unit 21, a second feature detection unit 22, a second feature collation unit 23, print character information 31, and an output unit 32 having a character recognition unit 32a for character recognition; Is provided. The character recognition unit 32a performs a well-known character recognition process.

  The form image input unit 1 is a processing unit that captures a form image via a camera, a scanner, or a communication line. The form image inclination correction unit 2 has a preset horizontal direction (hereinafter referred to as “x direction”) parallel to a horizontal ruled line of the form, and a predetermined reference vertical direction (hereinafter referred to as “x direction”). , “Y direction”) and a processing unit that corrects the inclination of the form image so that the vertical ruled lines of the form become parallel.

  The first edge image creation unit 11 is a first edge image creation processing unit that extracts a horizontal ruled line in a form image as an object. For example, a differential operator that detects an edge of a horizontal line (hereinafter referred to as “horizontal edge”). The horizontal edge is extracted by applying (for example, SOBEL operator) to the form image. Specifically, because there is a density difference between the horizontal ruled line and its upper part, there is a horizontal edge at the upper end of the horizontal ruled line, and there is a density difference between the horizontal ruled line and its lower part, so There are horizontal edges. Therefore, the first edge image creation unit 11 extracts two horizontal edges that form a pair with respect to one horizontal ruled line.

  The first edge image creation unit 11 binarizes each pixel of the differential image of the form image including the horizontal edge to create a first edge image composed of a binary image. In the binarization process, the binarization process can be performed using a threshold value set automatically using Otsu's discriminant analysis method or the like, or a predetermined threshold value set in advance. Here, a black pixel having a pixel value “0” indicates a pixel that forms an edge, and a white pixel that has a pixel value “1” indicates a pixel that does not form an edge. However, the processing can be performed so that the black pixel indicates a pixel that does not form an edge and the white pixel indicates a pixel that forms an edge.

  The first feature detection unit 12 is a processing unit that acquires a distribution pattern that is a feature amount of a horizontal ruled line in a form image. Specifically, the first feature detection unit 12 is included in the edge image created by the first edge image creation unit 11. A histogram in which the number of pixels at the edge of the horizontal ruled line is plotted in the y direction is acquired as a distribution pattern of the horizontal ruled line. The horizontal ruled line distribution pattern obtained in this way is normalized by compressing the entire distribution pattern to 50 dots / inch conversion.

  The first feature matching unit 13 compares the distribution pattern, which is the feature amount of the horizontal ruled line in the form image, with the first standard histogram 14 which is the distribution pattern of the horizontal ruled line of various forms registered in advance. It is a processing unit for calculating the degree.

  The second edge image creation unit 21 is a second edge image creation processing unit that extracts a vertical ruled line in a form image as an object. For example, a SOBEL that detects a vertical line edge (hereinafter referred to as “vertical edge”). A vertical edge is extracted by applying an operator or the like to the form image. The second edge image creation unit 21 binarizes each pixel of the differential image of the form image and creates a second edge image composed of a binary image.

  The second feature detection unit 22 is a processing unit that acquires a distribution pattern of vertical ruled lines in the form image. Specifically, a histogram in which the number of edge pixels of the vertical ruled lines is plotted in the y direction is used as the distribution pattern of the vertical ruled lines. get. The second feature matching unit 23 compares the distribution pattern of the vertical ruled lines in the form image with the second standard histogram 24, which is the distribution pattern of the vertical ruled lines of various forms registered in advance, and calculates the second matching degree. It is a processing unit.

  The print character information 31 is information about print characters printed on various forms registered in advance, and specifically includes coordinate information and print characters of the print characters printed on each form.

  The output unit 32 selects a plurality of leading candidate forms from various forms registered in advance using the first matching degree and the second matching degree, and print characters present at predetermined positions of the selected leading candidate forms This is a processing unit that outputs a form type having the highest character recognition score by comparing the result of reading from a predetermined position of the form image and comparing the result of character recognition by the character recognition unit 32a.

  Specifically, the output unit 32 selects a plurality of leading candidate forms in descending order of the total value of the first matching degree and the second matching degree, and prints the print characters on the leading candidate form from the print character information 31. Extract regions. Here, if there is a misalignment between the leading candidate form and the form image, the area where the print character is printed is corrected in consideration of this misalignment, and the data read from the corresponding part is recognized. . Then, the character recognized character is compared with the print character that should be originally included in the leading candidate form included in the print character information 31.

  In this embodiment, for the sake of convenience of explanation, a case will be described in which a print character serving as a headline printed on the top of a form is a processing target, but the present invention is not limited to this, and various forms It is also possible to process print characters located at various positions. In this case, information regarding a plurality of print characters may be included in the print character information 31 for each registered form. As the number of print characters to be processed increases, the reliability of the output result output from the output unit 32 can be improved.

  Next, the processing procedure of the form type discriminating apparatus 100 shown in FIG. 1 will be described. FIG. 2 is a flowchart showing processing of the form type discriminating apparatus 100 shown in FIG. In step S1 shown in the figure, the form image input unit 1 captures an image of the form. In step S2, the form image tilt correction unit 2 uses a known method so that the horizontal line segment in the form image matches the x direction, or the vertical line segment in the form image matches the y direction. As described above, the inclination angle of the form image is rotationally corrected. For example, the edge direction of the horizontal edge in the form image and the inclination angle in the x direction may be obtained, and affine transformation may be performed so as to rotationally correct this angle. Thereby, the horizontal ruled line in the form image can be matched with the x direction, and the vertical ruled line can be matched with the y direction.

  FIG. 4 illustrates a form image G after the inclination correction. In this form image G, the bold lines shown in the figure are ruled lines, the characters that are not italic are print characters that are printed in advance on the form, and the italic characters are characters that are entered on the form.

  Returning to FIG. 2, in step S <b> 3, the first edge image creation unit 11 applies a differential operator (e.g., SOBEL operator) that detects an edge forming a horizontal line to the form image to extract the horizontal edge, The pixel value of the horizontal edge is compared with a predetermined pixel value and binarized to create a first edge image that is a binary image as shown in FIG. Here, the black pixel of the first edge image indicates a pixel that forms a horizontal edge, and the white pixel indicates a pixel that does not form a horizontal edge.

  FIG. 5 illustrates the first edge image E1. As shown in the figure, when a differential operator for detecting a horizontal edge is applied to a ruled line (horizontal line) having a predetermined width, there are density differences between the upper and lower parts of the ruled line and the background. Parallel lines are detected. Further, since the differential operator for detecting the horizontal edge is applied, the vertical edge is not detected at this time. In addition, since there is a horizontal edge in a part of the printed character or entry character in the form image, the horizontal edge forming a part of the character is detected as shown in the figure. Are not detected in a continuous state.

  Returning to FIG. 2, in step S <b> 4, the first feature detection unit 12 compresses the first edge image so that 3 mm becomes one pixel in the x direction, and at this time, if a predetermined number of black pixels continue. If the pixel after compression is a black pixel, otherwise, the first compressed image is created by performing minimum value compression using the pixel after compression as a white pixel. FIG. 5 shows a first edge image. When the character is differentiated, only the horizontal edge is obtained, and further, the horizontal edge of the character disappears by the minimum value compression.

  FIG. 6 illustrates the first compressed image F1. When the width of the form image G is 210 mm and 1 pixel is formed per 1 mm, the number of pixels in the horizontal direction of the first compressed image F1 is 210/3 = 70. Note that when the first compressed image is created, the minimum value compression is performed in the horizontal direction as shown in FIG. 6, so the horizontal edge of the character seen in the first edge image E1 disappears.

  Returning to FIG. 2, in step S <b> 5, the first feature detection unit 12 creates a histogram in which the number of pixels of the horizontal edge of the first compressed image is plotted on the y coordinate, and so far compressed in the horizontal direction. A normalization process for compressing the whole in the vertical direction in terms of 50 dots / inch is performed, and the result wp is defined as the first histogram. FIG. 7 illustrates the first histogram H1, and FIGS. 8A to 8D illustrate various forms Ca to Cd and first standard histograms Ha to Hd thereof.

  Returning to FIG. 2, in step S <b> 6, the first feature matching unit 12 collates the first standard histogram and the first histogram H <b> 1 indicating the distribution of the horizontal edges of various forms registered in advance. Specifically, the first histogram H1 is divided into a plurality of divided areas having a predetermined number of pixels (for example, 20 pixels), and the position of each divided area of the first histogram H1 matches the first standard histogram Ha. And the first histogram Hy after correction and the first standard histogram Ha obtained by correcting each divided region so that the start position of the region coincides with the obtained position, and the first matching degree is obtained.

  This point will be described in more detail. As shown in FIG. 9, the first histogram H1 is divided into a plurality of partitioned areas A1 to A7. Here, for the partitioned area A1, a position on the first standard histogram Ha having an interval and a peak value between two peaks of the histogram in the partitioned area A1 is obtained. For example, the correlation value is sequentially obtained while moving the histogram in the partition area A1 on the first standard histogram H1, and the position where the correlation value is highest can be determined as the corresponding position of the partition area A1 in the first histogram H1.

  Similarly, regarding the partitioned areas A2 to A7, the correlation values are sequentially obtained while the histograms in the partitioned areas A2 to A7 are moved on the first standard histogram Ha, and the position where the correlation value is the highest is the partitioned area of the first histogram H1. It can be determined that the positions correspond to A2 to A7.

  In this way, when the positional deviation amounts y1 to y7 are obtained from the corresponding positions of the divided areas A1 to A7 shown in FIG. 10, the first histogram H1 is set to match the start position of the area with the obtained position. Correct for each A7. Thereafter, the correlation value between the corrected first histogram Hy and the first standard histogram Ha is obtained, and the first matching degree Pa between the form image G and the form Ca is calculated.

  Returning to FIG. 2, in step S <b> 7, the second edge image creation unit 21 extracts a vertical edge by applying a differential operator (e.g., SOBEL operator) that detects an edge forming a vertical line to the form image, The pixel value of each vertical edge is compared with a predetermined pixel value and binarized to create a second edge image that is a binary image.

  FIG. 11 illustrates the second edge image E2. Similarly to the first edge image E1 shown in FIG. 5, with respect to the second edge image E2, if a differential operator for detecting a vertical edge is applied to a ruled line (vertical line) having a predetermined width, Since there is a density difference between the left and right part and the background, two parallel lines are detected. Further, since the differential operator for detecting the vertical edge is applied, the horizontal edge is not detected. In addition, since a vertical edge exists in a part of the printed character or the written character in the form image, the vertical edge forming a part of the character is detected as shown in FIG. However, it is not detected as continuously as ruled lines.

  Returning to FIG. 2, in step S <b> 8, the second feature detection unit 22 compresses the second edge image so that 3 mm becomes one pixel in the y direction, and at this time, if the black pixels continue for a predetermined number of pixels. If the pixel after compression is a black pixel, otherwise, the second compressed image is created by performing minimum value compression with the pixel after compression being a white pixel. FIG. 12 illustrates the second compressed image F2. When the vertical width of the form image G is 300 mm and 1 pixel is formed per 1 mm, the number of pixels in the vertical direction of the second compressed image F2 is 300/3 = 100. Since the minimum value is compressed in the vertical direction, the characters on the form edge image E2 are lost.

  Returning to FIG. 2, in step S <b> 9, the second feature detection unit 22 creates a histogram of the number of black pixels in the y direction for each x coordinate of the second compressed image, and so far compressed in the vertical direction. The entire image is compressed in the direction of 50 dots / inch and used as the second histogram. FIG. 13 illustrates the second histogram H2, and FIGS. 14A to 14D illustrate various forms Ca to Cd and second standard histograms Ja to Jd for them.

  Returning to FIG. 2, in step S <b> 10, the second standard histogram indicating the distribution of the vertical edges of various forms registered in advance is compared with the second histogram H <b> 2. Specifically, the second histogram H2 is divided into a plurality of divided areas having a predetermined number of pixels (for example, 20 pixels), and the position of each divided area of the second histogram H2 matches the second standard histogram Ja. The corrected second histogram Hx obtained by correcting each partition region so as to match the obtained position is compared with the second standard histogram Ja to obtain the second matching degree.

  This point will be described in more detail. As shown in FIG. 15, the second histogram H2 is divided into a plurality of partitioned areas B1 to A4. Here, for the partitioned area B1, a position on the second standard histogram Ja having an interval and a peak value of two peaks of the histogram in the partitioned area B1 is obtained. For example, the correlation value is sequentially obtained while moving the histogram in the partitioned area B1 on the second standard histogram Ja, and the position where the correlation value is highest can be determined as the corresponding position of the partitioned area B1 in the second histogram H2.

  Similarly, regarding the partitioned areas B2 to B4, correlation values are sequentially obtained while moving the histograms in the partitioned areas B2 to B4 on the second standard histogram Ja, and the position where the correlation value is the highest is the partitioned area of the second histogram H2. It can be determined that the position corresponds to B2 to B4.

  In this way, when the positional shift amounts x1 to x4 are obtained from the corresponding positions of the divided areas B1 to B4 shown in FIG. 16, the second histogram H2 is divided into areas for the divided areas B1 to B4 so as to match the obtained positions. Correction for moving the start position coordinate value is performed. Thereafter, the correlation value between the corrected second histogram Hx and the second standard histogram Ja is obtained, and the second matching degree Qa between the form image G and the form Ca is calculated.

  Proceeding to step S11 in FIG. 3, the output unit 32 sets the order in which the total value of the first match degree and the second match degree is large in descending order of similarity.

  In step S12, for example, two are selected in descending order of similarity and are set as leading candidate forms. In step S13, one leading candidate form is selected from the leading candidate forms, and collation processing using print characters to be described later is performed. When selecting such a candidate candidate form, the first matching degree and the second matching degree are selected in descending order.

  In step S14, the coordinate information (vertical coordinate and abscissa) of the print character of the leading candidate form selected from information related to print characters of various forms registered in advance (hereinafter referred to as “print character information 31”) is read. . Specifically, in FIG. 17, the vertical area 15a (start point and end point) of the print character and the horizontal area 25a (start point and end point) of the print character are read out.

  In step S15, the vertical area 15a of the print character is moved in the reverse direction by the amount of positional deviation of the peak of the horizontal edge closest to the start point of the vertical area 15a of the read print character, and is the highest at the start point of the horizontal area 25a of the print character. The coordinate value of the horizontal area 25a of the print character is moved in the reverse direction by the amount of positional deviation of the peak of the near vertical edge. FIG. 18 shows a vertical region 25a ′ of the printed character moved in the reverse direction by the positional displacement amount y1 and a horizontal region 25a ′ of the printed character moved in the reverse direction by the positional displacement amount x1.

  In step S16, the characters in the area where the vertical area 15a 'of the moved print character and the horizontal area 25a' of the print character are applied to the form image are read by the character recognition process. FIG. 19 shows a region in which the vertical region 15a ′ of the moved region print character and the horizontal region 25a ′ of the print character are applied to the form image G. The character “payment slip” can be read from this area.

  In step S17, the first print character for the leading candidate form is read from the print character information 31 of various forms registered in advance. As shown in FIG. 17, the first print character of the form Ca is “payment slip”.

  In step S18, the area of the ordinate of the print character and the abscissa of the print character is extracted from the form image G, and it is determined whether or not the character in this area can be recognized as the first print character. Specifically, the area of the ordinate of the print character and the abscissa of the print character is extracted from the form image G, and when the character in this area can be recognized as the first print character, or the above-mentioned extracted from the form image G If there is no recognized character in the area and it is blank, it is determined that it matches the pre-registered print character information. As a result, if they match (step S18; Yes), the type of the strong candidate form is output (step S19), and the process ends.

  On the other hand, if it does not match the pre-registered print character information (step S18; No), it is determined in step S20 whether an unprocessed powerful candidate form exists, and an unprocessed powerful candidate form exists. If yes (step S20; Yes), the process proceeds to step S13 and the same process is repeated. And if the process with respect to all the promising candidate forms is complete | finished (step S20; No), a series of said process will be complete | finished.

According to the form type discriminating apparatus 100 of the present embodiment, the following effects can be obtained.
(A) A color image or monochrome image form image G is examined in the y direction to extract an appearing edge, and a first edge image E1 binarized by the presence or absence of an edge is created. On the first edge image E1, the horizontal ruled line appears as a horizontal parallel line separated by the ruled line width of the horizontal ruled line even if the density difference between the background and the ruled line is small, so that the horizontal ruled line can be extracted well.

(B) The first edge image E1 is compressed in the x direction to create a first compressed image F1 in which the number of pixels in the x direction is reduced. The number of black pixels continuous in the x-direction, such as written characters / graphics, pressed receipt / payment stamps, and pasted barcode labels, is not originally large on the first edge image E1. Since the number of black pixels continuous in the x direction, which is not originally large, is reduced by compression in the x direction, the presence on the first compressed image F1 becomes very sparse. On the other hand, the number of black pixels continuous in the x direction of the horizontal ruled line is not small on the first edge image E1. Even if the number of black pixels continuous in the x direction, which is not small, is reduced, the presence on the first compressed image F1 does not become dilute. Accordingly, the horizontal ruled line can be well extracted even in a form in which characters / graphics are entered, a cash stamp is pressed, or a barcode label is attached.

(C) The first edge image E1 is compressed to 3 pixels in the x direction to reduce the number of pixels in the x direction to reduce the number of pixels in the x direction, and the first compressed image F1 is created as a black pixel only when the predetermined number of pixels are continued. . Since most of the lengths of black characters that continue in the x direction for written characters are less than 10 mm, the first compressed image F1 has almost 0 to 3 pixels. Thereby, the presence of characters can be diluted. On the other hand, since most of the horizontal ruled lines in which the black pixels continue in the x direction are 10 mm or more, the first compressed image F1 has almost three pixels or more, and the presence of the horizontal ruled lines does not become dilute. Accordingly, the horizontal ruled line can be well extracted even in a form in which characters / graphics are entered, a cashier stamp is pressed, or a bar code label is attached.

(D) A color image or a monochrome form image is examined in the x direction to extract appearing edges, and a second edge image E2 binarized according to the presence or absence of edges is created. On the second edge image E2, vertical ruled lines can be successfully extracted because vertical ruled lines appear as vertical parallel lines that are separated by the ruled line width of the vertical ruled lines even if the form has a small density difference between the background and the ruled lines.

(E) The second compressed image F2 is generated by compressing the second edge image E2 in the y direction and reducing the number of pixels in the y direction. The number of black pixels that continue in the y direction, such as written characters / graphics, pressed receipt / payment stamps, and pasted barcode labels, is not originally large on the second edge image E2. Since the number of black pixels continuous in the y direction, which is not originally large, is reduced by compression in the y direction, the presence on the second compressed image F2 becomes very sparse. On the other hand, the number of black pixels continuous in the y direction of the vertical ruled line is not small originally on the second edge image E2. Even if the number of black pixels continuous in the y direction, which is not small, is reduced, the presence on the second compressed image F2 does not become dilute. Therefore, the vertical ruled line can be extracted well even in a form in which characters / graphics are entered, a cashier stamp is pressed, or a barcode label is attached.

(F) The second compressed image F2 is generated by compressing the second edge image by 3 mm in the y direction to 1 pixel to reduce the number of pixels in the y direction. Since the length of the black pixels continuing in the y direction for the written characters is almost less than 10 mm, the second compressed image F2 is almost 0 to 3 pixels. Thereby, the presence of characters can be diluted. On the other hand, the length of the vertical ruled lines in which the black pixels continue in the y direction is almost 10 mm or more, so that the second compressed image is almost 3 pixels or more, and the presence of the vertical ruled lines is not dilute. Therefore, the vertical ruled line can be extracted well even in a form in which characters / graphics are entered, a cashier stamp is pressed, or a barcode label is attached.

(G) Higher accuracy than before because it is possible to extract horizontal and vertical ruled lines with higher accuracy than before, even in forms with characters / figures filled in, stamps entered, and barcode labels attached. The type of form can be determined.

(H) Since the first histogram H1 is normalized to a size matching the first standard histograms Ha to Hd, it is easy to compare the two.
(I) Since the second histogram H2 is normalized to a size matching the second standard histograms Ja to Jd, it is easy to compare the two.

(J) Peaks in the first standard histograms Ha to Hd and the first histogram H1 indicate that there is a horizontal ruled line at the peak position. However, the form paper is caused by expansion / contraction of the form paper or deviation of cutting. The position of the horizontal ruled line may shift, and the peak position may shift even in the first histogram H1 related to the same type of form as the first standard histograms Ha to Hd. However, since the positional deviation amount of the plurality of horizontal ruled lines is the same for each horizontal ruled line, the corrected first deviation after correcting the positional deviation for each divided area so that the area start position matches the first histogram H1. By comparing the one histogram Hy with the first standard histograms Ha to Hd, it is possible to prevent the degree of matching from being lowered due to the shift of the horizontal ruled line position.

(K) The peaks in the second standard histograms Ja to Jd and the second histogram H2 indicate that there is a vertical ruled line at the peak position. However, the form paper is caused by expansion / contraction of the form paper or a shift in cutting. The position of the vertical ruled line may shift, and the peak position may shift even in the second histogram H2 related to the same type of form as the second standard histograms Ja to Jd. However, since the amount of positional deviation of the plurality of vertical ruled lines is the same for each vertical ruled line, the amount of positional deviation for each divided area is corrected so that the area start position matches the second histogram H2. By comparing the second histogram Hx and the second standard histograms Ja to Jd, it is possible to prevent the degree of matching from being lowered due to the shift of the vertical ruled line position.

(L) The y-direction position of the print area can be accurately determined based on the displacement amount in the y-direction obtained by comparing the peak positions in the first standard histograms Ha to Hd and the first histogram H1. Since the character printed in the print area is read to determine the type of form, the type of form can be determined with high accuracy.

(M) The position in the x direction of the print area can be accurately determined based on the amount of positional deviation in the x direction obtained by comparing the peak positions in the second standard histograms Ja to Jd and the second histogram H2. Since the character printed in the print area is read to determine the type of form, the type of form can be determined with high accuracy.

  The form type discriminating apparatus and form type discriminating method of the present invention can be used for an automatic form sorting apparatus.

DESCRIPTION OF SYMBOLS 1 Form image input part 2 Form image inclination correction part 11 1st edge image creation part 12 1st feature detection part 13 1st feature collation part 14 1st standard histogram of various forms 15 1st character area of various forms, 1st character 21 1st Two-edge image creation unit 22 Second feature detection unit 23 Second feature matching unit 24 Second standard histogram of various forms 31 Printed character information 32 Output unit 32a Character recognition unit 100 Form type discrimination device G Form image A1 to A7 Partition area B1 -B4 Partition area E1 First edge image E2 Second edge image F1 First compressed image F2 Second compressed image H1 First histogram H2 Second histogram Ha-Hd First standard histogram of various forms Ja-Jd Second of various forms Standard histogram

Claims (8)

Differential image generation means for generating a differential image obtained by differentiating the form image of the form in the vertical direction or the horizontal direction and binarizing;
A compressed differential image generating means for generating a compressed differential image obtained by compressing the differential image generated by the differential image generating means in the horizontal direction or the vertical direction;
Feature quantity generating means for generating a predetermined feature quantity related to the compressed differential image generated by the compressed differential image generating means;
A degree-of-match calculation means for calculating a degree of match using a feature amount of a compressed differential image relating to each form registered in advance and a feature amount generated by the feature amount generation means;
A form type discriminating apparatus comprising: output means for outputting a form type discrimination result based on the degree of match calculated by the match degree calculating means.   The compressed differential image generation means sets a pixel having a first predetermined length in the form as one pixel in the differential image generated by the differential image generation means, and the pixel having the first predetermined length. Of these, only when the second predetermined number of pixels are pixels that continuously form an edge portion, the minimum value compression is performed in the horizontal or vertical direction with the pixel value of the pixel assigned to the edge portion as the pixel value. The form type discriminating apparatus according to claim 1, wherein the compressed differential image is generated. The feature quantity generation means includes:
A histogram obtained by cumulatively adding the number of pixels forming the edge portion included in the compressed differential image in the horizontal direction or the vertical direction is generated as the feature amount.
The degree of coincidence calculation means calculates the degree of coincidence using a standard histogram of a compressed differential image relating to each form registered in advance and the feature amount generated by the feature amount generation means. 2. The form type discriminating apparatus described in 2.   The feature amount generation means normalizes a histogram obtained by accumulating the number of pixels forming the edge portion included in the compressed differential image in a horizontal direction or a vertical direction to a size matching the standard histogram. The form type identification device according to claim 3, further comprising:   The degree-of-match calculation means obtains a position where the peak position in the standard histogram matches the peak position in the histogram for each divided area set in the standard histogram of various forms, and matches the obtained position. 5. The form type discriminating apparatus according to claim 3, wherein the histogram is corrected, and the degree of coincidence is calculated by comparing the corrected histogram with a standard histogram.   The output means obtains a corresponding area on the form image corresponding to a print area set in association with the form type having a high degree of matching based on the amount of positional deviation, and extracts characters in the corresponding area. 6. The form type discriminating apparatus according to claim 5, wherein the form type that can be recognized as the print character set in association with the form type having a high degree of coincidence is output. Differential image generation means for generating a first differential image obtained by differentiating the form image of the form in the vertical direction and binarizing and a second differential image obtained by differentiating the form image of the form in the horizontal direction and binarized;
A first compressed differential image obtained by compressing the first differential image generated by the differential image generating means in the horizontal direction and a second compressed differential image obtained by compressing the second differential image generated by the differential image generating means in the vertical direction. A compressed differential image generating means for generating;
Feature amount generating means for generating a first feature amount relating to the first compressed differential image generated by the compressed differential image generating means and a second feature amount relating to the second compressed differential image generated by the compressed differential image generating means;
A degree-of-match calculation means for calculating a degree of match using the first feature quantity and the second feature quantity relating to each form registered in advance and the first feature quantity and the second feature quantity generated by the feature quantity generation means;
A form type discriminating apparatus comprising: output means for outputting a form type discrimination result based on the degree of match calculated by the match degree calculating means. A differential image generating step of generating a differential image obtained by differentiating the form image of the form in the vertical direction or the horizontal direction to binarize;
A compressed differential image generating step for generating a compressed differential image obtained by compressing the differential image generated by the differential image generating step in the horizontal direction or the vertical direction;
A feature amount generating step for generating a predetermined feature amount related to the compressed differential image generated by the compressed differential image generating step;
A degree-of-match calculation step of calculating a degree of match using the feature amount of the compressed differential image for each form registered in advance and the feature amount generated by the feature amount generation step;
A form type discrimination method comprising: an output step of outputting a form type discrimination result based on the degree of match calculated in the match level calculation step.

Images