JP2006119942A - Character string area segmenting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a character string area segmenting device which segments a character string area from a document image of an arbitrary format. <P>SOLUTION: The device includes a display means 110 for displaying the document image, input means 120 and 130 each inputting coordinates of a character string area designation point 152 pointing an objective character string area 102 of document images 101 and 151 displayed on the display means, and a boundary detection means 140 for detecting the character string area from the document images on the basis of coordinates of the character string area designation point. Thus the character string area can be easily and accurately segmented. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image processing technique for a document image, and more particularly to a character string area cutout device that cuts out a character string area to be subjected to character recognition.

  Devices equipped with character recognition technology, such as an optical character reader (OCR), are widely used. An application of OCR is electronic processing or management of document images.

  There are various types of document images, such as bank transfer forms and postcards, forms such as business cards, and free description documents such as papers and articles once printed on paper. As a specific example of electronic document management, the character recognition result of the candidate number on the answer sheet of the exam is used as the management number, and the answer sheet is stored and managed as an image. The character recognition result of the title name of the academic paper is searched. As an example, a search is performed.

  In general character recognition technology, character recognition processing is first performed on the entire document. In the above example, only the necessary character string may be cut out and character recognition processing performed. When the document image is an image of a form or a free description document, a method of cutting out a necessary character string from the document image is well known (see, for example, Patent Document 1 and Patent Document 2). In Patent Document 1, when the document image is a form image, the form format definition such as “There is a column of 2 centimeters in each position in this position on the screen” and “the content written in the column is a monetary amount”. There is a description of a form processing method in which the knowledge of a form is held in the apparatus in advance and a necessary character string can be cut out by cutting out a column corresponding to the required character string. In Patent Document 1, in the case of a free description document, a character string to be clipped is partially scanned, or the entire document image is displayed on a display device, a range to be clipped by a mouse operation is designated, and a scan range Alternatively, there is a description of an image processing system that extracts a necessary character string by setting a designated range as a cutout range.

  The process of selecting a similar color pixel to be connected by clicking on a pixel in the image and cutting it out is not limited to a form document or a free description document. It can be set manually whether it is included (for example, refer patent document 3).

JP 2003-323553 A

JP 2000-174945 A JP 2000-353239 A

  The technique described in Patent Document 1 has a problem that it takes time and effort to input the form knowledge to the apparatus, and it can be applied only to forms that hold the form knowledge in the apparatus. In the technique described in Patent Document 2, there is a problem that the character string region cannot be accurately cut out unless the scanning or range specification is strictly performed, and both simplicity and accuracy cannot be achieved. In the technique described in Patent Document 3, there is a problem that even if the characters are the same, such as the hiragana of hiragana, it is not possible to cut out unconnected portions as the same lump.

  An object of the present invention is to provide a character string area segmentation device that can accurately segment a target character string area from a document image in an arbitrary form format.

  The character string region cutout device of the present invention is a region cutout device that cuts out a character string region including a lexical phrase and / or a symbol surrounded by blank pixels (white pixels) from a document image. A character string area cutout device according to the present invention includes a display means for displaying a document image, and at least one character string area designation point for indicating a target character string area of the document image displayed on the display means. Input means for inputting coordinates and boundary detection means for detecting a character string area from a document image based on the coordinates of a character string area designation point. The boundary line detecting means detects the coordinates of the boundary line surrounding the character string region from the document image based on the coordinates of the character string region designated point. The character string region designation point is at least one arbitrary point in the character string region, or at least one arbitrary point in the vicinity of the character string region. The boundary line detection means detects a boundary line surrounding the character string area.

  According to the apparatus of the present invention, since the character string area is cut out using only the pixels around the character string area designation point, it is automatically and accurately performed without requiring knowledge of a form for a document image in an arbitrary form format. Can be cut out.

  In the following embodiments, a document image is described as being a binary image that is composed of white pixels (blank pixels) and black pixels, and valid information is displayed with black pixels.

  The character string region cutout device of the present invention is a region cutout device that cuts out a character string region composed of characters surrounded by white pixels and / or symbols including blank symbols from a document image. The document image may include an image data area.

  A character string area cutout device according to the present invention inputs a display means for displaying a document image, and an area designation point for designating a target character string area and / or image data area of the document image displayed on the display means. And a boundary detection unit for detecting a character string region from the document image based on the coordinates of the region designation point. A typical example of a document image is a receipt.

  A pixel that designates a character string (character string area designation point) is an arbitrary pixel in the character string area, an arbitrary pixel in which the closest black pixel is in the character string area, or a black pixel closest in the direction of the character string Is an arbitrary pixel in the character string area, and is designated by the operator through one or more input operations of click, gesture, horizontal line, rectangular box, and circle box. The boundary line detection means detects a character string area.

  Based on the coordinates of the character string region designated point, the boundary line detection means scans the pixels of the document image in the order close to the character string region designated point and detects the coordinates of the boundary line surrounding the character string region from the document image. The coordinates of the boundary line are any of the coordinates of the curve, two points that specify a rectangle having four right corners, and four points that specify a quadrilateral having four corners other than a right angle.

  FIG. 1 is a diagram showing an outline of a configuration example of an apparatus according to an embodiment of the present invention. The apparatus according to the embodiment of the present invention includes a display unit 110 that displays a document image, and character string region designation points 152 that indicate a target character string region 102 of the document images 101 and 151 displayed on the display unit. Input means 120 and 130 for inputting coordinates, and boundary line detection means 140 for detecting a character string area from a document image based on the coordinates of a character string area designation point, and from a document image in an arbitrary form format, The character string area can be cut out easily and accurately.

  In FIG. 1, a personal computer 141 as an arithmetic processing unit includes, as input means, for example, a mouse 120, a keyboard 130, a display means 110, a storage means for storing a document image 101 (a binary image with white pixels and black pixels), For example, a boundary line detection unit 140 configured by software (program) that performs automatic calculation for cutting out a character string region “Hello” from the document image 101 is included. The document image 101 input from the boundary line detection unit 140 is displayed as a document image 151 displayed on the display unit 110. The document image 151 is a binary image composed of white pixels and black pixels.

  In FIG. 1, as an example of the document image 101, character strings “IROHA”, “NI”, “HOHOTO”,..., “ABC” horizontally written on each line indicated by each item number 1 to 5 are displayed. 110. In the document image 101, an area in which the character string “Honoto” of item number 3 that the user (operator) wants to cut out is referred to as a “character string area”, and one point above the character “to”. One point that the user inputs to the device in order to designate a character string area to be cut out by the user is called a “character string area designation point”.

  The user looks at the document image 151 and designates the character string area designation point 152 using the mouse 120 or the keyboard 130. The boundary line detection unit 140 receives the character string area designation point 152, automatically detects the coordinates of the boundary line from the character string area designation point 152 and the stored document image 101, and the character string determined by the boundary line. A region 102 (shown as a binary image with white and black pixels) 102 is cut out.

  FIG. 2 is a flowchart showing an outline of the procedure of region extraction in the embodiment of the present invention.

  The area cut-out includes a document image display step (step) 210 for displaying a document image on a display unit, a character string region designation point input step (step) 220 for a user to input a character string region designation point using the input unit, and a document image. And a boundary line detection step (step) 230 for automatically cutting out the character string area from the document image by the boundary line detection means from the character string area designation point.

  FIG. 3 is a diagram for explaining an example of the character string region designation point input step 220 shown in FIG.

  FIG. 3 shows a part 310 (shown as a binary image with white pixels and black pixels) of the document image displayed on the display unit 110, and the user views the document image 151 displayed on the display unit 110. Look at it, click the mouse 120, and enter a character string area designation point 320, which is one point on the character string area "Hoto" to be cut out. In the boundary line detection step 230, for example, a character string region is automatically cut out from a document image based on a character string region designation point by a combination of expansion processing and contour extraction processing. If dilation processing is performed only in the vicinity of the pixel to be scanned in the contour extraction processing, the same cutout result as that obtained when the contour extraction processing is performed after dilating the entire image can be obtained, and the number of pixels to be scanned can be reduced. In the expansion process, the black pixel is expanded by changing the four pixels closest to the black pixel to the black pixel, and in the contour extraction process, a process of detecting the boundary between the description in the image and the background is performed. .

  FIG. 4 is a diagram showing an example of a character string region cut out in the boundary line detection step 230 shown in FIG. In FIG. 4, the thick line indicates the boundary line 410 of the extracted character string region, and indicates a pixel group 420 (consisting of white pixels and black pixels) scanned by the contour extraction process in order to extract the character string region from the document image. ing.

  According to the apparatus of the embodiment of the present invention described above, since the boundary line of the character string area is automatically detected only by the user specifying one point, the target character string area can be easily and accurately detected. Can be cut out. Further, since the character string area is cut out using only the image around the character string area designation point, it is possible to cut out the document image of any form format without requiring knowledge of the form.

  FIG. 5 is a diagram showing an example of a method in which the user inputs a character string region designated point using an input means such as a mouse in the character string region designated point input step 220 shown in FIG.

  In addition to the click described in FIG. 3, the input method of the character string region designation point is any one of horizontal lines, gestures, rectangular boxes, and round boxes shown in FIGS. 5 (a) to 5 (d). FIG. 5 shows a part 510 of a document image, a drag start point 520, a drag end point 530, a character string region designation point 540 designated as a result of the drag, and drag trajectories 551 and 552. In the case of the horizontal line shown in FIG. 5A, the middle point of the drag start point 520 and the drag end point 530 is set as a character string region designation point 540. In the case of the gesture shown in FIG. 5B, the point indicated by the gesture is set as a character string region designation point 540. Here, it is assumed that which one of the gesture and the gesture corresponds to the character string region designation point is determined in advance. In the case of the rectangular box shown in FIG. 5C, the center of gravity of the rectangle is set as the character string area designation point 540. In the case of the circle shown in FIG. 5D, the center of gravity of the circle is set as the character string region designation point 540.

  As shown in FIG. 5, in the apparatus according to the embodiment of the present invention, since the character string region designation point 540 can be input by various input methods, high operability can be obtained using an optimal input method.

  Next, a configuration in which a character string to be cut out can be accurately cut out even when the user cannot accurately operate the mouse or the like and a point deviated from the character string to be cut out is input to the character string region designation point is shown in FIGS. 7 for explanation.

  FIG. 6 is a flowchart showing an outline of a region cutout procedure having a character string region designated point correction step (step) 610 in the embodiment of the present invention.

  FIG. 7 is a diagram showing an example of processing of the character string region designated point correction step 610 shown in FIG.

  6 adds a character string region designation point correction step 610 for automatically correcting the character string region designation point input in the character string region designation point input step 220 according to a certain rule to the configuration shown in FIG. The structure which performs the boundary line detection process 230 with respect to the result of the character string area | region designation | designated point correction process 610 is shown. FIG. 7 shows a part 710 of a document image (consisting of white pixels and black pixels). In the following description, it is assumed that a character string including the character “ho” is cut out from the part 710 of the document image. As shown in FIG. 7, points deviated from the character string to be cut out, for example, white pixels (black pixels constituting the character “ho”, for example, pixels other than the black pixel 720) 730 and 740 are character string region designation points. Can be automatically corrected according to certain rules.

  The correction rule for correcting the character string area designation point is "If the input character string area designation point is a white pixel, the black pixel closest to this white pixel is used as the character string area designation point". For example, the pixel 720 which is the closest black pixel from the white pixel 730 becomes a corrected character string region designation point, and the black pixel 720 includes the character “ho” because it is a black pixel constituting the character “ho”. A character string can be cut out. In addition, the correction rule for correcting the character string region designation point is as follows: `` If the input character string region designation point is a white pixel, the black pixel present at the shortest distance from this white pixel in the direction of the character string If the character string area designation point is “the character string area designation point”, the black pixel 720 present at the shortest distance in the character string direction from the white pixel 740 becomes the corrected character string area designation point, and the character including the character “ho” A column can be cut out. When the black pixel 720 is searched from the white pixel 740 and a character string including “ho” is cut out, the distance between the black pixel 720 and, for example, the black pixel 750 in the same column as the black pixel 720 is compared. Pixel 720 is selected. In the above description, the character string direction is described as the horizontal horizontal direction, but the character string direction may be any direction. When associating the character string region designation point with the character string to be cut out using the distance in the character string direction, it is assumed that the character string direction is given in advance or has been detected in advance by some means.

  As described above, in the apparatus according to the embodiment of the present invention, in addition to the black pixel 720, as exemplified by the white pixels 730 and 740, a plurality of character string region designation points for the character string to be cut out are illustrated. Therefore, there are few restrictions regarding input, such as “the user needs to specify a character string region designation point with black pixels while avoiding white pixels”. Therefore, according to the apparatus of the embodiment of the present invention, even when the user cannot accurately operate the mouse or the like and a point deviated from the character string to be cut out is input to the character string area designated point, the character string to be cut out is cut out. be able to. That is, the user can easily input the character string region designation point.

  An example in which a plurality of character string areas are detected and output by a boundary line detection process when a document image and a character string area designation point for the document image are given will be described below with reference to FIGS.

  FIG. 8 is a diagram illustrating a definition example of neighboring pixels in the boundary line detection step 230 in the embodiment of the present invention.

  FIG. 9 is a diagram showing an example of detection and output in the boundary line detection step 230 in the embodiment of the present invention.

  As shown in FIG. 9, a plurality of character string regions can be detected for the same document image and character string region designation point, and an output can be obtained. The expansion process in the boundary line detection step described with reference to FIG. 3 is a process of changing the four white pixels (four neighboring pixels) near the target pixel to black pixels when the target pixel is a black pixel. As shown in FIG. 8, a neighboring pixel 820 indicated by hatching that is uniquely defined in place of the four neighboring pixels is used for the target pixel 810. The neighboring pixels defined by FIG. 8A, FIG. 8B, and FIG. 8C are respectively shown in FIG. 8D, FIG. 8E, and FIG. The neighboring pixels defined by (f) are respectively designated by parameters p = 1, 2, and 3, and the neighboring pixels defined by FIGS. 8 (g), 8 (h), and 8 (i) are designated by parameter p. = 1, 2, and 3, if managed in order, a plurality of character string regions ordered with respect to the same document image and character string region designation point can be detected and output.

  For example, as shown in FIG. 9A, when a character string region designation point 910 is input to the document image 101, the neighborhood pixel definition (p = 1) 921 is set as shown in FIG. 9B. On the other hand, the character string region 931, the definition of neighboring pixels (p = 2) 922 for the definition of the character string region 932, the definition of neighboring pixels (p = 4) the definition 923, the character string region 933, the definition of neighboring pixels (p = 6) A character string area 934 is detected and output for 924. The character string regions 931 to 934 are composed of white pixels and black pixels. The neighborhood pixel definition (p = 1) 921 is used to cut out a single character, and the neighborhood pixel definition (p = 2) 922 is used to cut out a character string having a narrow character interval. ) 924 is suitable for extracting a character string having a wide character interval such as a character string including an item number.

  The boundary detection process described in FIGS. 8 and 9 allows the apparatus of the embodiment of the present invention to cut out a plurality of ordered character string regions. For example, the plurality of character string regions are displayed on the screen. By leaving the final judgment regarding the region extraction to the user, the accuracy of the region extraction can be improved.

  In the description using FIGS. 1 to 4 and FIGS. 8 to 9, the example of realizing the boundary detection process by combining the expansion process and the contour extraction process is shown. Another method, for example, a process of characterizing the target pixel by the number of the target pixel and the number of black pixels near the target pixel may be used instead of the expansion process. Characterizing is, for example, processing in which the target pixel is black if the number of black pixels is 1 or more, and white if the number of black pixels is 0.

  FIG. 10 is a flowchart showing an outline of the procedure for segmentation having a step of automatically setting the definition of the optimum neighboring pixel for each image document in the embodiment of the present invention.

  FIG. 10 shows the form issuing agency described in the configuration shown in FIG. 2 in the font size of the print character, the character string direction such as vertical writing or horizontal writing, or the document image other than the character string to be cut out. A neighborhood definition setting step (step) 1010 for setting the definition of neighboring pixels using a description to be specified or a description for designating the form type is added, and the definition of the neighboring pixels set by the neighborhood definition setting step 1010 is used to create a boundary. The structure which performs the line detection process 230 is shown.

  In the neighborhood definition setting step 1010, for example, a neighborhood pixel definition (p = 1) 921 is set for a character string with a narrow character spacing, and a neighborhood pixel definition (p = 6) 924 is set for a wide character spacing. This setting is automatically set according to a preset document type. Alternatively, it is automatically set according to the recognition result of other areas in the document image. For example, the number of the form issuing organization is read at the top of the document image, and the definition of the neighboring pixels is set according to the form format of the issuing organization.

  The apparatus of the embodiment of the present invention has the neighborhood definition setting step shown in FIG. 10 and can select one of the detection results using other information related to the document image, for example, character spacing. It is possible to automatically obtain an appropriate cutout result.

  In the apparatus of the embodiment of the present invention, the coordinates representing the boundary line are the coordinates of the curve, the coordinates of four points that specify a figure having right angles at the four corners, and the coordinates of four points that specify a figure having angles other than a right angle at the four corners. The coordinates of two points that specify a rectangle having right angles at the four corners, or the coordinates of four points that specify a quadrilateral (typically, a rhombus or a trapezoid) having an angle other than a right angle at the four corners may be used.

  FIG. 11 is a diagram illustrating an example of coordinates representing a character string area (consisting of white pixels and black pixels) in the embodiment of the present invention. 11A shows coordinates 1110 for linearly approximating a curve, FIG. 11B shows two-point coordinates 1121 and 1122 for specifying a rectangle having four corners of 90 degrees, and FIG. 11C has four corners of 90 degrees. An example of four-point coordinates 1131, 1132, 1133, 1134 for specifying a figure is shown.

  As shown in FIG. 11, even if the detected boundary line is a curved line, the apparatus according to the embodiment of the present invention uses coordinates that approximate the curve linearly, coordinates of two points and four points that represent the figure, that is, A character string area can be expressed with a small amount of information. Therefore, there is no inconvenience when the cut character string area is passed to the character recognition engine, and the apparatus of this embodiment can be incorporated as a component of another apparatus without any trouble.

  FIG. 12 is a diagram showing an example of a receipt that is a target of the apparatus according to the embodiment of the present invention. FIG. 13 is an example of a document image including an embedded character string that is a target of the apparatus according to the embodiment of the present invention. FIG. 14 is a diagram showing an example of a document partially including a diagram, which is a target of the apparatus according to the embodiment of the present invention. In addition to the examples shown in FIGS. 12, 13, and 14, it goes without saying that the object of the apparatus of the embodiment of the present invention has many application examples.

  FIG. 12 shows an example of a receipt image 1210, which is a form for a medical institution to describe a medical practice and expenses and charge the health insurance association or the like for medical expenses. The receipt 1210 includes a patient symbol / number 1220 ( In the example of the figure, “insurance 1234567”, “symbol / number 1234567”), and the treatment prescription name (“diastase” in the example of the figure) 1230 which is a drug name and a test name are described. The apparatus according to the embodiment of the present invention cuts out the patient's symbol / number 1220, makes the patient's symbol / number 1220 the management number, and can perform electronic document management documents by character recognition and search. In addition, it is possible to support the task of checking the appropriateness of the treatment prescription by cutting out the drug name 230 and performing character recognition and searching for the efficacy and the maximum prescription amount for each drug.

  The character strings described in the figures are often important phrases such as conclusions, summaries and keywords derived from the figures. FIG. 13 shows a character string 1320 described in the figure, for example, conclusion, summary, etc. (in the example, “2.7 times difference between Kochi Prefecture (1st place) and Kanagawa Prefecture (47th place)”) An example of a document image 1310 in which is embedded is shown. With the apparatus according to the embodiment of the present invention, an important word / phrase in the drawing can be extracted from such a document image by searching, and an electronic document management document can be obtained.

  FIG. 14 shows an example of a document 1410 that partially includes a figure or picture area (image data area), and the apparatus according to the embodiment of the present invention designates an important figure from such a document. The image data area 1420 can be cut out. 13 and 14 can be realized by changing the definition of the neighboring pixels illustrated in FIG.

  In the above-described embodiments, the document image is composed of white pixels (blank pixels) and black pixels, and the valid information is a binary image displayed with black pixels. However, the valid information is white. Needless to say, the present invention can be similarly applied to a binary image displayed by pixels.

  In the embodiment described above, the document image is described as a binary image composed of white pixels and black pixels. However, a threshold value is set for the pixels of the multi-value image to binarize the multi-value image. By obtaining a binary image, the present invention can also be applied to a multi-valued image.

  According to the apparatus of the present invention, a character string region can be automatically cut out by designating one point of coordinates, so that a necessary character string can be cut out easily and accurately. In addition, since the character string area can be cut out using only the image around the character string area designation point, it is possible to cut out a document image in an arbitrary form format without requiring knowledge of the form.

  It is possible to provide a character string area cutout device that can automatically cut out a target character string area from a document image in an arbitrary form format.

The figure which shows the outline | summary of the structural example of the apparatus of the Example of this invention. The flowchart which shows the outline | summary of the procedure of area | region extraction in the Example of this invention. The figure explaining an example of the character string area | region designation | designated point input process shown in FIG. The figure which shows the example of the character string area | region cut out by the boundary line detection process shown in FIG. The figure which shows the example of the method in which a user inputs a character string area | region designated point in the character string area | region designated point input process shown in FIG. The flowchart which shows the outline | summary of the procedure of area extraction which has a character string area | region designated point correction process in the Example of this invention. The figure which shows the example of a process of the character string area | region designated point correction process shown in FIG. The figure which shows the example of a definition of the near pixel in the boundary line detection process in the Example of this invention. The figure which shows an example of the output of the boundary line detection process in the Example of this invention. The flowchart which shows the outline | summary of the procedure of area extraction which has the process of setting automatically the definition of the optimal neighborhood pixel for every image document in the Example of this invention. The figure which shows an example of the coordinate expressing a character string area | region in the Example of this invention. The figure which shows the example of the receptor which is the object of the apparatus of the Example of this invention. The figure which shows the example of the document image containing the embedded character string which is the object of the apparatus of the Example of this invention. The figure which shows the example of the document which is a target of the apparatus of the Example of this invention, and contains a figure in part.

Explanation of symbols

DESCRIPTION OF SYMBOLS 110 ... Display means, 120 ... Mouse, 130 ... Keyboard, 140 ... Boundary line detection means, 141 ... Personal computer, 102 ... Character string area, 101, 151 ... Document image, 152, 320, 540, 910 ... Character string area designation Point 210 ... Document image display step 220 ... Character string region designation point input step 230 ... Boundary line detection step 310, 510, 710 ... Part of document image 410 ... Boundary line of clipped character string region 420 ... Pixel group scanned by contour extraction processing, 520 ... Drag start point, 530 ... Drag end point, 551, 552 ... Drag trajectory, 610 ... Character string region designation point correction step, 730, 740 ... White pixels, 720, 750 ... black pixel, 810 ... target pixel, 820 ... neighboring pixel, 921, 922, 923, 924 ... definition of neighboring pixel, 931 ... near Character string region for pixel definition 921, 932... Character string region for definition of neighboring pixel 922, 933... Character string region for definition of neighboring pixel 923, 934... Character string region for definition of neighboring pixel 924, 1010. Step 1110: Coordinates for approximating a curve in a straight line, 1121, 1122: Two points for designating a rectangle having four corners of 90 degrees, 1131, 1132, 1133, 1134 ... Four points for designating a figure having four corners of 90 degrees 1210 ... Receipt image, 1220 ... Patient symbol / number, 1230 ... Treatment prescription name, 1310 ... Document image in which character string is embedded, 1320 ... Character string, 1410 ... Document including figures and pictures, 1420 ... Designation Image data area.

Claims (9)

  Display means for displaying a document image; input means for inputting coordinates of a character string area designating point indicating a target character string area of the document image displayed on the display means; and the character string area designation A character string region segmentation device comprising boundary line detecting means for detecting the character string region from the document image based on the coordinates of a point.   2. The character string region cutout device according to claim 1, wherein the boundary line detection unit scans pixels of the document image in an order close to the character string region designation point to detect a boundary line surrounding the character string region. A character string region segmentation device characterized by the above.   2. The character string region cutout device according to claim 1, wherein the character string region designation point is specified by an operator by one or more input operations of click, gesture, horizontal line, rectangular box, and circle box. A character string region segmentation device characterized by being the center of gravity of a point or range.   2. The character string region cutout device according to claim 1, wherein the character string region designation point is an arbitrary pixel in the character string region, an arbitrary pixel in which a nearest black pixel is in the character string region, or a character string A character string region cutout device, wherein a black pixel closest to the direction of any one of the pixels in the character string region is any pixel.   3. The character string region segmentation device according to claim 2, wherein the coordinates of the boundary line are coordinates of a curve, two points for designating a rectangle having four right corners, and four points for designating a quadrilateral having four corners other than right angles. A character string region cutout device characterized in that any of the coordinates is used.   2. The character string region cutout device according to claim 1, wherein the boundary line detection unit detects a plurality of the character string regions.   7. The character string region cutout device according to claim 6, wherein the character space described in a region other than the character string region to be cut out of the document image, a form issuing organization is designated, and the form type is designated. A character string region segmentation device that selects and detects one of the plurality of character string regions based on one or more of them.   2. The character string region cutout device according to claim 1, wherein the document image includes a description frame, and the coordinates of the character string region designation point are the coordinates of the upper, lower, left, and right edges of the description frame. Row area segmentation device. The character string region segmentation device according to claim 1, wherein the document image is a receipt.

Images