JP2011146028A - Character recognition method and character recognition device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a character recognition method and a character recognition device, which exactly delimit a concatenation character. <P>SOLUTION: The character recognition device identifies one or more candidate division paths at each candidate division point in a character image which is a recognition target, divides the character image which is the recognition target based on each of identified candidate division paths, and generates one or more image segments. Then, the character recognition device recognizes each of one or more image segments and obtains one or more recognition results. Then, the character recognition device selects one or more best division points and best division path of each best division point of the character image from the candidate division points and the candidate division paths at candidate division points, based on the obtained recognition results. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to the field of character recognition, and more particularly to a character recognition method and a character recognition device.

  Conventionally, when character segmentation is inaccurate, it becomes one of the causes of errors in subsequent character recognition. Therefore, research on character segmentation is important in character recognition.

  For example, a character recognition device that divides an input image including a plurality of characters into images for each character is disclosed. Specifically, the character recognition device recognizes a connected part in the input image by the recognition logic, classifies the connected part recognized by the classification logic, and the connected part is an image of one character according to the classification result. It is determined whether or not there is. When the character recognition device is not an image of one character, the connected portion is divided again into smaller segments by the separation logic, or the adjacent segments are merged by the merge logic. The character recognition device repeatedly executes the classification, segmentation, and merge processing until the division result satisfies a predetermined condition.

  Also, several character image segmentation methods are disclosed. These include White Space and Pitch Analysis, Projection Analysis, Connected Component Processing, Character Division by Recognition, Mixed Division, and Whole Recognition. .

US Pat. No. 5,787,194

Richard G. Casey, Eric Lecolinet, “A Survey of Methods and Strategies in Character Segmentation”, IEEE Transactions on Pattern Analysis and Machine Intelligence, July 1996, Vol. 18, No. 7, p. 690-706

  However, the above-described conventional technique has a problem that it is impossible to accurately separate connected characters.

  For example, it is difficult to specify an accurate division position and division path in the separation of connected characters such as handwritten characters, italic type printed on a device, ornamental type, or cursive type.

  Specifically, in the process of segmenting a character image, the character recognition device uses a segmentation position between characters (hereinafter referred to as a segmentation point) and a segmentation path for each segmentation position in order to segment a plurality of characters one by one. (Hereinafter referred to as a split path). For example, the character recognition device divides each character in the character image by specifying one division path for each division point. FIG. 13A is a diagram illustrating an example in which the character recognition device specifies one division path for each division point. However, in many cases, particularly in the process of separating connected characters (for example, handwritten characters, italic or decorative characters printed on devices, cursive letters, etc.), there may be a plurality of division paths that can be used at each division point. There is sex. FIG. 13B is a diagram illustrating an example in which the character recognition device identifies one or a plurality of possible division paths at each of a plurality of possible division points in the character image. Whether these possible division points (hereinafter referred to as candidate division points) and their possible division paths (hereinafter referred to as candidate division paths) are correct directly affects the accuracy of character recognition. In a known method, the character recognition device first specifies one division path for each division point. Thereafter, the character recognition device determines whether or not the image cut in this way satisfies a predetermined criterion. If the predetermined standard is not satisfied, the character recognition device performs division or merge again until the predetermined standard is satisfied. That is, the character recognition device again specifies each of the division points and one corresponding division path. However, the division points and the division paths specified by such a method only satisfy a predetermined standard, and are not optimal division points and optimum division paths in many cases.

  The disclosed technology has been made in view of the above, and an object thereof is to provide a character recognition method and a character recognition device capable of accurately separating connected characters.

  For a basic understanding of some aspects of the disclosed technology, an overview of the disclosed technology is provided below. This summary is not an exhaustive list of disclosed technology. This is not intended to establish key points or important parts of the disclosed technology, nor is it intended to limit the scope of the disclosed technology. The purpose is to give the concept of the disclosed technology in a concise form. This summary is a preface to the more detailed description below.

  According to one aspect of the disclosed technology, a character recognition method is provided. This method specifies one or a plurality of candidate division paths for each candidate division point in the character image to be recognized (wherein at least one candidate division point has a plurality of candidate division paths), Based on each candidate division pass, the character image to be recognized is divided to generate one or more image segments, and one or more image segments are recognized to recognize one or more Obtaining a result, and selecting one or a plurality of optimum division points of the character image and an optimum division path for each optimum division point from the candidate division points and the candidate division paths at the candidate division points based on the recognition result. .

  According to one aspect of the disclosed technology, a character recognition device is provided. The apparatus identifies one or a plurality of candidate division paths for each candidate division point in the character image to be recognized (here, at least one candidate division point has a plurality of candidate division paths). Recognizing each of one or more image segments, a specific unit, an image segment generation unit that divides a character image to be recognized based on each candidate division path, and generates one or more image segments A recognition unit that acquires one or a plurality of recognition results, and one or more optimum division points and optimum division points of the character image from the candidate division points and the candidate division paths at the candidate division points based on the recognition results. And a path search unit for selecting the optimal divided path.

  The embodiment of the disclosed technique further provides a computer program for realizing the character recognition method.

  Note that the embodiments of the disclosed technology further provide a computer program product in which computer program codes for realizing a character recognition method are recorded in at least a computer-readable medium format.

  According to one aspect of the character recognition method and the character recognition device disclosed in the present application, it is possible to accurately separate connected characters.

FIG. 1 is a flowchart schematically illustrating a character recognition method according to the first embodiment. FIG. 2 is a flowchart schematically illustrating another example of the character recognition method according to the first embodiment. FIG. 3 is a flowchart schematically illustrating the character recognition method according to the second embodiment. FIG. 4A is a diagram illustrating an example in which candidate division points are marked with a first marker. FIG. 4B is a diagram illustrating an example in which candidate division paths are marked with a second indicator. FIG. 5 is a flowchart schematically showing an example in which the optimum division path is searched by applying dynamic programming. FIG. 6 is a flowchart schematically illustrating the character recognition method according to the third embodiment. FIG. 7A is a diagram illustrating an example of another division. FIG. 7B is a diagram illustrating an example of merger. FIG. 8 is a flowchart schematically illustrating the character recognition method according to the fourth embodiment. FIG. 9 is a block diagram schematically illustrating a character recognition apparatus according to the fifth embodiment. FIG. 10 is a block diagram schematically illustrating a character recognition apparatus according to the sixth embodiment. FIG. 11 is a block diagram schematically illustrating a character recognition device according to the seventh embodiment. FIG. 12 is a block diagram schematically illustrating an example of a computer that can be applied to an embodiment of the disclosed technology. FIG. 13A is a diagram illustrating an example in which the character recognition device specifies one division path for each division point. FIG. 13B is a diagram illustrating an example in which the character recognition device identifies one or a plurality of possible division paths at each of a plurality of possible division points in the character image.

  Hereinafter, embodiments of the disclosed technology will be described with reference to the drawings. Elements and features described in one drawing or type of embodiment of the disclosed technology may be combined with elements and features shown in one or more other drawings or embodiments. In the drawings and the description, in order to clarify the disclosed technique, display and description of components and processes that are not related to the disclosed technique and are known to those skilled in the art are omitted.

  Referring to the description of the embodiments of the disclosed technology in combination with the drawings, the objects, features, and effects of the disclosed technology can be understood more easily. The components in the drawings merely illustrate the principles of the disclosed technology. In the drawings, the same or similar technical features and parts are denoted by the same or similar reference numerals.

  FIG. 1 is a flowchart schematically illustrating a character recognition method according to the first embodiment. As illustrated in FIG. 1, the character recognition device searches for an optimal division path from one or a plurality of candidate division paths for each candidate division point by executing the processes of steps S203 to S209.

  In step S203, the character recognition device identifies one or a plurality of candidate division paths for each candidate division point in the character image to be recognized. Here, it is assumed that among a plurality of candidate division points in the character image to be recognized, a plurality of candidate division paths exist at at least one candidate division point.

  Here, the character image to be recognized may be in any appropriate format. Although not all are enumerated here, the character image to be recognized may be, for example, a binary image, a JPG image, a bitmap image, or the like. In addition, the character image can include any type of characters such as numerals, symbols, and letters or characters of various pronunciations. Here, no limitation is imposed on the character image.

  In addition, the character recognition device may specify candidate division points and candidate division paths using any appropriate method. For example, character recognition devices include white space and pitch analysis, projection analysis, connected component processing, and ascender-descender analysis. ) Etc. can be used. Here, detailed description of these existing methods is omitted.

  The character recognition device detects all possible division paths for each candidate division point as candidate division paths in order to identify the optimum division path.

  In step S205, when the candidate division path is specified at each candidate division point, the character recognition device divides the character image that is the recognition target based on the candidate division paths, and generates one or a plurality of image segments. To do.

  In step S207, the character recognition device recognizes each of the generated one or more image segments and acquires one or more recognition results.

  In step S207, the character recognition device may recognize the image segment using any appropriate method. FIG. 2 is a flowchart schematically illustrating another example of the character recognition method according to the first embodiment. Step S703 shown in FIG. 2 is the same as step S203 shown in FIG. Step S705 is the same as step S205. Step S709 is the same as step S209. Therefore, these descriptions are omitted here. In step S707, the character recognition device recognizes an image segment. The character recognition device executes sub-step S707-1 and sub-step S707-2 in this step. In substep S707-1, the character recognition device classifies the generated one or more image segments. The character recognition device classifies a plurality of image segments generated using any appropriate pattern recognition method according to the type of character in the image. For example, the character recognition device uses a neural network, Bayesian decision, support vector machine, structural pattern recognition method, etc., and a combination of these methods. Details of these methods will not be described in detail here. In sub-step S707-2, the character recognition device calculates a matching distance between each type of image segment and the type of known character image as a recognition result of the image segment. For example, if a certain image segment is classified as belonging to the letter “a” after classification, the character recognition device calculates a matching distance between the image segment and the known image of the letter “a”. To do. Step S707 in FIG. 2 shows only one example of recognizing an image segment, but the disclosed technique is not limited to this example. In addition, the character recognition apparatus may use any appropriate recognition technology such as neural network, Bayesian determination, support vector machine, structural pattern recognition method, and a combination of these methods. Detailed description of these will be omitted.

  A description will be given using an example in which a plurality of corresponding known images exist for a certain character type. When a certain image segment is classified into the type, the character recognition device calculates a matching distance between the image segment and each of the plurality of known images. After that, the character recognition device selects a minimum matching distance as a recognition result corresponding to the image segment, calculates an average value or a total value thereof for the matching distance, and calculates the calculated value The recognition result corresponding to the character image is used.

  Finally, in step S209, based on the recognition result, the character recognition device determines one or a plurality of optimum division points of the character image from the candidate division points and the candidate division paths at the candidate division points and the optimum division point for each optimum division point. Select a path.

  For example, since there may be a plurality of candidate division paths at a candidate division point of a character in the character image to be recognized, in step S205, the character recognition device selects a plurality of image segments corresponding to the character. Generate. Accordingly, in step S207, the character recognition device acquires a plurality of recognition results corresponding to these image segments. For example, in the example of FIG. 2, the character recognition device calculates a plurality of matching distances for a plurality of image segments corresponding to a certain character. Here, the smaller the calculated matching distance, the better the recognition result. Therefore, the character recognition device selects an image segment corresponding to the minimum matching distance. In other words, the character recognition device can select an optimal division point and an optimal division path at the optimal division point from candidate division points that can be obtained by dividing the image segment and candidate division paths.

  In the above-described method, the character recognition device does not have to determine one determination criterion in advance. That is, the character recognition device can specify one or a plurality of candidate division paths at one division point. Therefore, since the character recognition apparatus can select the optimum division path from one or a plurality of candidate division paths, the accuracy of character recognition can be greatly improved.

  The method for selecting the optimum division point and the optimum division path of the character image from the plurality of candidate division points and one or more candidate division paths is not limited to this example. In the second embodiment, the character recognition device marks each candidate division path for each candidate division point, and uses a dynamic programming algorithm from a plurality of candidate division points and one or a plurality of candidate division paths. The optimum division point and the optimum division path at the optimum division point are selected.

  FIG. 3 is a flowchart schematically illustrating the character recognition method according to the second embodiment. As shown in FIG. 3, in this method, the character recognition apparatus executes the processes of steps S303, 304, 305, 307, and 309.

  Since step S303 is the same as step S203 in the first embodiment shown in FIG. 1, the description thereof is omitted here.

  In step S304, the character recognition device marks each candidate division path with a two-dimensional label for each candidate division point. As described above, the character recognition apparatus marks a plurality of image segments by the two-dimensional labeling of these candidate division paths. The two-dimensional mark includes a first mark for marking each candidate dividing point and a second mark for marking each candidate dividing path. Specifically, the character recognition device includes I (I> 1) candidate division points in the character image to be recognized, and J (1 ≦ i ≦ I) -th candidate division point J ( A case where J ≧ 1) candidate division paths are specified will be described as an example. The character recognition device marks the j-th (1 ≦ j ≦ J) -th candidate division path of the i-th candidate division point as (i, j) or (j, i). Thus, the image segment obtained by carving out by the j-th candidate division pass of the i-th candidate division point and the n-th candidate division pass at the m (1 ≦ m ≦ I) -th candidate division point. Is represented as {(i, j), (m, n)} or {(j, i), (n, m)}. It is assumed that N candidate division paths exist on the m-th division point, and N ≧ 1, 1 ≦ n ≦ N.

  FIG. 4 shows an example of performing two-dimensional marking. The image shown in FIG. 4A includes a handwritten word “net” as a character image to be recognized, and candidate division points specified therein include “position i”, “position i + 1”, and “position i + 2”. Including. Here, i ≧ 0. The candidate split path at position i can be labeled (i, 1), and the candidate split path at position i + 1 can be labeled (i + 1,1). Then, each of the candidate division paths at the position i + 2 can be labeled as (i + 2, 1), (i + 2, 2), and (i + 2, 3). Thus, in FIG. 4, the image segment obtained by carving out using the candidate division path (i, 1) at the position i and the candidate division path (i + 1, 1) at the position i + 1 is {(i, 1), (I + 1,1)}.

  Since step S305 is the same as step S205 in the first embodiment shown in FIG. 1, the description thereof is omitted here. Further, step S307 is the same as step S207 in the first embodiment shown in FIG.

  In step S309, the character recognition device selects an optimal division point and an optimal division path from a plurality of candidate division points and candidate division paths at the candidate division points based on the recognition result acquired in step S307. For example, the character recognition device uses the dynamic programming method based on the two-dimensional label, and the optimal division point and the optimal division point at the optimal division point from the candidate division point and the candidate division path at the candidate division point. Search for a path.

  FIG. 5 is a flowchart schematically showing an example in the case of searching for an optimal division path using dynamic programming.

  As shown in FIG. 5, after two-dimensional marking is performed on a plurality of image segments, a dynamic programming method is used to identify a plurality of candidate division points and candidate division paths at each candidate division point. One or a plurality of optimum dividing points and an optimum dividing path for each optimum dividing point are selected. In this method, the character recognition device executes the processes of steps S1209-1, 1209-2, and 1209-3.

  First, in step S1209-1, the character recognition device classifies a plurality of image segments. Specifically, the character recognition device classifies the plurality of image segments for each of the left (or right) division points based on the candidate division points corresponding to the image segments and the two-dimensional markers of the candidate division paths. Here, image segments having the same candidate division point on the left side (or right side) are classified for each candidate division path.

  Next, in step S1209-2, the character recognition device defines a connection relationship between the image segments. Specifically, the character recognition device defines the connection relationship between each image segment according to the following procedure, that is, the character recognition device has a candidate division point and a candidate division path on the right side of one image segment, Only when the candidate division point on the left side of the other image segment is the same as the candidate division path, the two image segments are defined as connected. The character recognition device determines whether or not the candidate division points corresponding to the two image segments and the candidate division paths are the same using the two-dimensional label.

  Next, in step S1209-3, the character recognition device searches for a path. Specifically, the character recognition device detects the candidate division path corresponding to the optimum recognition result as the optimum division path of the image using dynamic programming based on the connection relation between the defined image segments. The candidate division points corresponding to these are set as the optimum division points. As an example, a plurality of corresponding known images may exist for a certain character type. The character recognition device calculates a matching distance between each of the plurality of known images for each image segment, and uses an average value or a total of the plurality of matching distances as a recognition result of the image segment. For example, the character recognition device detects a divided path with the shortest matching distance using dynamic programming. Here, the character recognition device determines that the image segment constituting the divided path with the shortest matching distance has the best recognition result.

  In the above-described method, the character recognition device searches for the optimum division path of the character image using dynamic programming, so that the matching distance can be calculated efficiently.

  In Example 3, the character recognition apparatus identifies each candidate division point and each candidate division path of the character image to be recognized, and then merges and / or divides the obtained image segment again. FIG. 6 is a flowchart schematically illustrating the character recognition method according to the third embodiment. As shown in FIG. 6, in this method, the character recognition apparatus executes the processes of steps S603, 605, 606, 607, and 609.

  Step S603 is the same as step S203 in the first embodiment. Step S605 is the same as step S205 in the first embodiment. Therefore, description of step S603 and step S605 is omitted here.

  In step S606, the character recognition apparatus selectively performs re-division and / or merge on the plurality of image segments obtained in step S605. Specifically, the character recognition device may merge adjacent image segments and / or may divide one or more of the plurality of image segments again. Thereby, the character recognition device generates a new image segment.

  FIG. 7 shows an example of performing another division and / or merge. Here, FIG. 7A shows an example of re-division, and FIG. 7B shows an example of merging. FIG. 7A shows a case where two image segments, that is, “image segment 1” and “image segment 2” in FIG. 7A are obtained after step S605. The character recognition device can perform the division again on the two image segments. In this example, the character recognition device performs division again on the image segment 2 to obtain the image segment 3 and the image segment 4. In FIG. 7B, the character recognition apparatus obtains four image segments, that is, “image segment 1′-4 ′” in FIG. 7B after step S605. The character recognizer can merge with adjacent segments in these image segments. For example, the character recognition device acquires the image segment 5 ′ by merging the image segment 1 ′ and the image segment 2 ′. As another example, in step S606, the character recognition device may perform re-division on one or a plurality of image segments at the same time as merging the adjacent image segments.

  In step S607, the character recognition device recognizes all the image segments obtained in steps S605 and S606. For example, in the example of FIG. 7A, the character recognition device recognizes the obtained four image segments 1-4 and acquires corresponding recognition results. In the example of FIG. 7B, the character recognition device recognizes the obtained five image segments (image segments 1′-5 ′) and acquires corresponding recognition results. Note that step S607 for obtaining the recognition result is the same as step S207 in the first embodiment, and a description thereof will be omitted here.

  Since step S609 is the same as step S209 in the first embodiment, description of step S609 is omitted here.

  In the above-described method, the character recognition device can divide the image segment again and / or merge to prevent erroneous division of the image segment and accurately search for the optimum division path.

  FIG. 8 is a flowchart schematically illustrating the character recognition method according to the fourth embodiment. As shown in FIG. 8, in the method, the character recognition device executes a process of recognizing a connected part in step S801.

  In step S801, the character recognition device recognizes one or more connected portions in the character image to be recognized. For example, it is assumed that the input image is an image including the handwritten character “chorghui@xxx.net” shown in FIG. The character recognition device divides the handwritten character “chorghui@xxx.net” into four connected parts including “chorghui @”, “xxx”, “.”, And “net”. The character recognition apparatus may recognize the connected portion by using any appropriate method, for example, a space and pitch analysis method, a commonly used connected portion analysis method, or an ascender / descender analysis method. In addition, detailed description about recognition of a connection part is abbreviate | omitted here.

  The character recognition device executes each process for these connected portions. In step S803, the character recognition device specifies and marks each candidate division point for each connected portion. In step S804, the character recognition device identifies and marks one or more candidate division paths for each candidate division point. Step S803 is similar to step S303 in the second embodiment illustrated in FIG. 3, and step S804 is similar to step S304. In other words, in steps S803 and S804, the character recognition device specifies one or a plurality of candidate division paths for each candidate division point in the connected portion, and for each candidate division path by two-dimensional labeling. Do the title. The marking method is the same as that in the second embodiment, and the description thereof is omitted here.

  In step S805, the character recognition device generates a plurality of image segments by performing division on each of the connected portions using the candidate division points and candidate division paths identified in steps S803 and S804. .

  In step S806, the character recognition apparatus performs another division and / or merge on the plurality of image segments acquired in step S805. Specifically, the character recognition apparatus can merge adjacent image segments and / or perform another division of one or more of the plurality of image segments. In this way, the character recognition device generates a new image segment.

  In the fourth embodiment, the character recognition apparatus can further mark the new image segments. For example, in the example shown in FIG. 7A, the character recognition device marks a division point where the image segment 2 is divided again as a position i + 2. On the other hand, the character recognition apparatus marks the corresponding candidate division path as (i + 2, j). Here, 1 ≦ j ≦ J, where J is the number of candidate division paths at the division point. In the example shown in FIG. 7B, the character recognition apparatus marks {(i, 0), (i + 2,0)} as the image segment 5 ′ obtained by merging the image segment 1 ′ and the image segment 2 ′. The left boundary of the image segment 5 'indicates the 0th candidate division path of the i-th division point, and the right boundary indicates the 0th candidate division path of the i + 2th division point. In this way, the character recognition device completes the marking of each merged image segment.

  In step S807, the character recognition device recognizes all the image segments acquired in steps S805 and 806, and acquires a recognition result. The method for acquiring the recognition result is the same as that in step S207 in the first embodiment, and the description thereof is omitted here.

  In step S809, the character recognition device uses the dynamic programming method based on the two-dimensional label according to the recognition result acquired in step S807, and determines the optimum division point and the candidate division point from the plurality of candidate division points and candidate division paths. Search for the optimal split path.

  In step S811, the character recognition device determines whether or not processing for all connected portions has been completed. If the processes for all the connected parts are not completed, the character recognition apparatus returns to step S803 and executes the subsequent processes. If the character recognition device has completed the processing for all the connected portions, the character recognition device ends the processing.

  In this embodiment, the character recognition apparatus performs processing on each of the connected portions in the input image, so that the processing on the input image can be made more efficient.

  FIG. 9 is a block diagram schematically illustrating a character recognition apparatus 900 according to the fifth embodiment. As illustrated in FIG. 9, the character recognition device 900 includes a candidate path specifying unit 902, an image segment generation unit 904, a recognition unit 906, and a path search unit 908.

  The candidate path specifying unit 902 specifies one or a plurality of candidate division paths for each candidate division point in the character image to be recognized, and outputs the candidate division points and the candidate division paths to the image segment generation unit 904. Here, it is assumed that at least one candidate division point among a plurality of candidate division points of the image has a plurality of candidate division paths.

  As in the first to fourth embodiments, the character image to be recognized may have any appropriate format. For example, the character images to be recognized are binary images, JPG images, bitmap images, etc., but not all are listed here. The character image can include any type of character, for example, numbers, symbols, various pronunciation words or characters, and is not limited in any way here. In addition, the candidate path specifying unit 902 specifies a candidate division point and a candidate division path using any appropriate method. For example, the candidate path specifying unit 902 includes white space and pitch analysis, projection analysis, connected component processing, and ascender-descender analysis. Descender Analysis) is used. Note that detailed description of these methods is omitted here.

  The image segment generation unit 904 divides the character image to be recognized using the candidate division path specified by the candidate path specification unit 902, generates one or a plurality of image segments, and generates a plurality of image segments. The data is output to the recognition unit 906.

  The recognition unit 906 recognizes each of one or more image segments received from the image segment generation unit 904 and acquires one or more recognition results. Then, the recognition unit 906 outputs the recognition result to the path search unit 908.

  The recognition unit 906 recognizes an image segment using the method described in the first embodiment. In addition, description is abbreviate | omitted here.

  Based on the recognition result received from the recognition unit 906, the path search unit 908 determines one or more candidate division points and one or more optimum division points and optimum division points of the character image from the candidate division paths at the candidate division points. Select the optimal split path for each.

  The path search unit 908 can specify a plurality of candidate division paths at one division point of the character image without needing to set a determination criterion in advance, and selects an optimum division path from the plurality of identified candidate division paths. To do. Therefore, the character recognition device 900 can greatly improve the accuracy of character recognition.

  FIG. 10 is a block diagram schematically illustrating a character recognition apparatus 1000 according to the sixth embodiment. A character recognition device 1000 illustrated in FIG. 10 includes a candidate path specifying unit 1002, an image segment generation unit 1004, a recognition unit 1006, and a path search unit 1008. Here, the candidate path specifying unit 1002 has the same function as the candidate path specifying unit 902 in the embodiment shown in FIG. Further, the image segment generation unit 1004 has the same function as the image segment generation unit 904 in the embodiment shown in FIG. Therefore, description of the candidate path specifying unit 1002 and the image segment generating unit 1004 is omitted here.

  The difference from the embodiment shown in FIG. 9 is that the recognition unit 1006 has a classification unit 1006a and a matching unit 1006b. The classification unit 1006a classifies one or a plurality of image segments generated by the image segment generation unit 1004. The classification unit 1006a uses the classification method described in the first embodiment. Note that description of the classification method is omitted here. Based on the classification result output from the classification unit 1006a, that is, based on the type of each image segment, the matching unit 1006b calculates the matching distance between the type of known character image for each image segment. Calculated as a recognition result. For example, when the image is classified by the classification unit 1006a, the matching unit 1006b, when an image segment is classified as belonging to the letter “a”, the matching distance between the image segment and the known image of the letter “a”. Is calculated. For example, a plurality of corresponding known images may be stored in the recognition unit 1006 for a certain character type. When a certain image segment is classified into the type by the classification unit 1006a, the matching unit 1006b calculates a matching distance between the image segment and each of the plurality of known images. The matching unit 1006b selects the calculated minimum matching distance as a recognition result corresponding to the image segment. Furthermore, the matching unit 1006b calculates an average value or a total value thereof for the matching distance, and uses the calculated value as a recognition result corresponding to the image segment.

  The path search unit 1008 determines the optimum division point and optimum division path of the character image from the candidate division paths at the plurality of candidate division points and the complementary division points according to the recognition result output from the recognition unit 1006, that is, the calculated matching distance. Select. Specifically, for a certain character in a character image, there may be a plurality of candidate division paths at the corresponding candidate division point, so that a plurality of image segments corresponding to the character in the image segment generation unit 1004. May be generated. In response to this, the recognition unit 1006 may obtain a plurality of recognition results corresponding to these image segments. For example, for a plurality of image segments corresponding to a certain character, the matching unit 1006b calculates a plurality of matching distances. Note that the smaller the calculated matching distance, the better the recognition result. Therefore, the path search unit 1008 can select an image segment corresponding to the minimum matching distance. That is, the path search unit 1008 selects a division path obtained by dividing the image segment as the optimum division path at the corresponding division point.

  Furthermore, as an example, the candidate path specifying unit 902/1002 may mark each of one or a plurality of candidate division paths for each candidate division point with a two-dimensional label. Here, the two-dimensional marker includes a first marker for marking each candidate segmentation point and a second marker for marking each candidate segmentation path. The candidate path specifying unit 902/1002 can use the marking method described in the second embodiment. Note that description of the marking method is omitted here. Based on the two-dimensional label marked by the candidate path specifying unit 902/1002, the path search unit 908/1008 uses dynamic programming to optimize the candidate division points and candidate division paths at the candidate division points. A division point and an optimal division path may be selected. Here, the path search unit 908/1008 may perform a path search using the dynamic programming described in the second embodiment. Note that description of path search using dynamic programming is omitted here. The path search unit 908/1008 can improve processing efficiency by searching for a character split path using dynamic programming.

  In another example, the image segment generation unit 904/1004 divides an image, merges adjacent image segments, and / or applies to one or more of the plurality of image segments. Then perform another division. Then, the image segment generation unit 904/1004 may output all the obtained image segments to the recognition unit 906/1006. The image segment generation unit 904/1004 can perform division and / or merge again on the generated image segment using the method described in the third and fourth embodiments, and the description is omitted here. . When the image segment generation unit 904/1004 performs the division and / or merge again on the image segment, the character recognition device can prevent the image segment from being erroneously divided, and accurately searches for the optimum division path. be able to.

  FIG. 11 is a block diagram schematically illustrating a character recognition device 1100 according to the seventh embodiment. Similar to the character recognition device 1000 in the sixth embodiment, the character recognition device 1100 illustrated in FIG. 11 includes a candidate path identification unit 1102, an image segment generation unit 1104, a recognition unit 1106, and a path search unit 1108. The difference is that the character recognition device 1100 further includes a connected portion specifying unit 1101.

  The connected part specifying unit 1101 recognizes one or more connected parts in the character image to be recognized, and outputs the obtained connected part to the candidate path specifying unit 1102. Since the connection part specifying unit 1101 operates using the same method as in the fourth embodiment, the description thereof is omitted here.

  The candidate path specifying unit 1102 processes each of the connected parts specified by the connected part specifying unit 1101. The candidate path specifying unit 1102 specifies one or a plurality of candidate division paths for each candidate division point for each connected portion. Further, the candidate path specifying unit 1102 may mark each candidate division path with a two-dimensional marker. In addition, since the method of description is the same as the method demonstrated in Example 2, description is abbreviate | omitted here.

  The image segment generation unit 1104 divides each connected portion using the candidate division points and candidate division paths identified by the candidate path identification unit 1102, and generates a plurality of image segments. As an example, the image segment generation unit 1104 can perform division and / or merge again on these image segments. Specifically, the image segment generation unit 1104 can merge adjacent image segments and / or execute re-division on one or more of the plurality of image segments. You can also Therefore, the image segment generation unit 1104 generates a new image segment. The image segment generation unit 1104 can also mark these new image segments. Since the title method is the same as that of the fourth embodiment described with reference to FIG. 8, the description thereof is omitted here.

  The recognition unit 1106 recognizes all the image segments generated by the image segment generation unit 1104, and acquires corresponding recognition results. The recognition of the image segment may be processed by the same method as that described in the first to fourth embodiments, and thus the description thereof is omitted here. The path search unit 1108 selects the optimum division point and the optimum division path of the character image from the plurality of candidate division points and the candidate division paths according to the recognition result received from the recognition unit 1106. A specific search method may be processed in the same manner as the method described in the first to fourth embodiments. For example, the path search unit 1108 uses dynamic programming based on the two-dimensional signs marked by the candidate path identification unit 1102 and the image segment generation unit 1104 according to the recognition result acquired by the recognition unit 1106. . As a result, the path search unit 1108 selects an optimal division point and an optimal division path from a plurality of candidate division points and candidate division paths.

  In the seventh embodiment, the character recognition device 1100 can process the input image more efficiently by processing each connected portion in the input image.

  Each example described in the examples in this specification is an example, and the disclosed technology is not limited thereto.

  Each component in the character recognition device described above may be arranged as software, firmware, hardware, or a combination thereof. Specific techniques or schemes that can be used for the arrangement are well known to those skilled in the art, and will not be described here. When implemented by software or firmware, a program constituting the software is installed from a storage medium or a network to a computer having a dedicated hardware configuration (for example, the general-purpose computer 1300 shown in FIG. 12). When the various programs are installed, the general-purpose computer 1300 executes various functions.

  In FIG. 12, the central processing unit (CPU) 1301 executes various processes by a program stored in a read-only memory (ROM) 1302 or a program uploaded from the storage unit 1308 to a random access memory (RAM) 1303. To do. The RAM 1303 also stores data necessary for the CPU 1301 to execute various processes as necessary. The CPU 1301, ROM 1302, and RAM 1303 are connected via a bus 1304. Similarly, the input / output interface 1305 is connected to the bus 1304.

  The input unit 1306, the output unit 1307, the storage unit 1308, and the communication unit 1309 are connected to the input / output interface 1305. Here, the input unit 1306 includes a keyboard, a mouse, and the like. For example, the output unit 1307 includes a display such as a cathode ray tube (CRT) or a liquid crystal display (LCD), a speaker, and the like. The storage unit 1308 includes a hard disk and the like. For example, the communication unit 1309 includes a network interface card such as a LAN card, a modem, and the like. The communication unit 1309 executes communication processing via a network, for example, the Internet. A drive 1310 may be connected to the input / output interface 1305. For example, a removable medium 1311 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is attached to the drive 1310, and a computer program read from the medium 1311 is installed in the storage unit 1308.

  When the above-described series of processing is executed by software, the general-purpose computer 1300 is installed with a program constituting the software from a network or a storage medium. Here, the network is, for example, the Internet, and the storage medium is, for example, a removable medium 1311.

  Such a storage medium is not limited to the removable medium 1311. For example, the storage medium may be a ROM 1302, a hard disk included in the storage unit 1308, or the like. A program is stored therein and distributed to users together with a device including the program. The removable medium 1311 stores a program therein as shown in FIG. 12, and provides the program to the user by being distributed away from the device. For example, the removable medium 1311 includes a magnetic disk (including a floppy (registered trademark) disk), an optical disk (including a compact disk read-only memory (CD-ROM) and a digital versatile disk (DVD)), and a magneto-optical disk (including Including a mini disk (registered trademark) and a semiconductor memory.

  The disclosed technology further provides a program product in which an instruction code readable by a device is stored. When the instruction code is read and executed by the device, the method of the embodiment according to the disclosed technology can be executed.

  On the other hand, the disclosed technology also includes a storage medium on which a program product in which an instruction code readable by a device is stored. The storage medium includes, but is not limited to, a floppy (registered trademark) disk, an optical disk, a magneto-optical disk, a memory card, a memory stick, and the like.

  In the above description of specific examples of the disclosed technology, an example of the embodiment is described, and the characteristics shown as an example of the embodiment are the same or similar, and one or more other embodiments. May be used. Further, it may be combined with features in other embodiments, or may be replaced with features in other embodiments.

  Where the term “including / having” is used in the text, it means the presence of a feature, element, step or component, but the presence or absence of one or more other features, elements, steps or components. It is supplemented not to exclude addition.

  The method of the disclosed technology is not limited to being executed in the order of explanations described in the specification, and may be executed simultaneously or independently according to other orders. Therefore, the execution order of the method described in this specification does not limit the technical scope of the disclosed technology.

  As described above, the disclosed technology has been described by describing a specific embodiment of the disclosed technology. However, the above-described embodiment is merely an example and is not limiting. Those skilled in the art can design various modifications, improvements, or similar products within the scope of the technical idea of the disclosed technology. These modifications, improvements, or similar products should be considered as being within the scope of protection of the disclosed technology.

  The following supplementary notes are further disclosed with respect to the embodiments including the above examples.

(Supplementary note 1) The character recognition device
Identifying one or more candidate division paths for each candidate division point in the character image to be recognized;
Dividing the character image to be recognized based on each candidate division path to generate one or a plurality of image segments;
Recognizing each of the one or more image segments to obtain one or more recognition results;
Selecting one or a plurality of optimum division points of the character image and an optimum division path for each optimum division point from the candidate division points and the candidate division paths at the candidate division points based on the recognition result. A character recognition method characterized by the above.

(Appendix 2) The character recognition device
After executing the step of identifying candidate division paths for each candidate division point, a two-dimensional indicator including a first indicator for marking each candidate division point and a second indicator for marking each candidate division path The character recognition method according to claim 1, further comprising the step of marking each of one or a plurality of candidate division paths for each candidate division point using.

(Supplementary Note 3) Step of selecting one or a plurality of optimum division points of the character image and an optimum division path for each optimum division point from the candidate division points and candidate division paths at the candidate division points based on the recognition result Is
Including a step of searching for an optimum division point and an optimum division path for each optimum division point from the candidate division points and the candidate division paths at the candidate division points by dynamic programming based on the two-dimensional label for each candidate division path. 2. The character recognition method according to 2.

(Supplementary Note 4) Recognizing the one or more image segments and obtaining one or more recognition results includes:
Classifying the one or more image segments;
The character recognition method according to claim 1, further comprising: calculating a matching distance between each type of image segment and the type of known character image as a recognition result of the image segment based on the type of the image segment.

(Supplementary note 5) The character recognition device
Before performing the step of recognizing the one or more image segments and obtaining one or more recognition results, merge adjacent image segments and / or one of the plurality of image segments. The character recognition method according to appendix 1, further comprising executing a step of dividing again one or more.

(Appendix 6) The character recognition device
Before executing the step of identifying one or more candidate division paths for each candidate division point in the character image to be recognized, one or more connected portions in the character image to be recognized are recognized. The character recognition method according to claim 1, further comprising executing the step.

(Supplementary Note 7) A candidate path specifying unit that specifies one or a plurality of candidate division paths for each candidate division point in the character image to be recognized;
An image segment generation unit that divides the character image to be recognized by each candidate division pass and generates one or a plurality of image segments;
A recognition unit that recognizes each of the one or more image segments and obtains one or more recognition results;
A path search unit that selects one or a plurality of optimum division points of the character image and an optimum division path for each optimum division point from the candidate division points and candidate division paths at the candidate division points based on the recognition result; A character recognition device comprising:

(Supplementary Note 8) The candidate path specifying unit further includes:
One or more candidate divisions for each of the candidate division points using a two-dimensional marker that includes a first indicator for marking each candidate division point and a second indicator for marking each candidate division path The character recognition device according to appendix 7, which is arranged so as to mark each of the paths.

(Supplementary Note 9) The path search unit further includes:
The character recognition device according to appendix 8, wherein the character recognition device is arranged so as to search for an optimum division point and an optimum division path from the candidate division points and the candidate division path based on the two-dimensional mark by dynamic programming.

(Supplementary Note 10) The recognition unit
A classifier for classifying the one or more image segments;
The character recognition device according to appendix 7, further comprising: a matching unit that calculates a matching distance between the type of each image segment and the known character image of the type for each image segment as a recognition result of the segment of the image. .

(Supplementary Note 11) The image segment generation unit further includes:
Arranging adjacent image segments and / or performing re-division on one or more of the plurality of image segments and outputting the resulting image segments to the character recognition device The character recognition device according to appendix 7.

(Additional remark 12) The character recognition apparatus of Additional remark 7 further provided with the connection part specific | specification part which recognizes the 1 or several connection part in the character image which is recognition object.

900 Character recognition device 902 Candidate path identification unit 904 Image segment generation unit 906 Recognition unit 908 Path search unit 1000 Character recognition device 1002 Candidate path identification unit 1004 Image segment generation unit 1006 Recognition unit 1006a Classification unit 1006b Matching unit 1008 Path search unit 1100 Character Recognizing device 1101 Linked part specifying unit 1102 Candidate path specifying unit 1104 Image segment generating unit 1106 Recognizing unit 1108 Path searching unit 1300 General-purpose computer 1301 CPU
1302 ROM
1303 RAM
1304 Bus 1305 Input / output interface 1306 Input unit 1307 Output unit 1308 Storage unit 1309 Communication unit 1310 Drive 1311 Removable medium

Claims (10)

The character recognition device
Identifying one or more candidate division paths for each candidate division point in the character image to be recognized;
Dividing the character image to be recognized based on each candidate division path to generate one or a plurality of image segments;
Recognizing each of the one or more image segments to obtain one or more recognition results;
Selecting one or a plurality of optimum division points of the character image and an optimum division path for each optimum division point from the candidate division points and the candidate division paths at the candidate division points based on the recognition result. A character recognition method characterized by the above. The character recognition device
After executing the step of identifying candidate division paths for each candidate division point, a two-dimensional indicator including a first indicator for marking each candidate division point and a second indicator for marking each candidate division path The character recognition method according to claim 1, further comprising the step of marking each of one or a plurality of candidate division paths for each candidate division point using. Based on the recognition result, selecting one or a plurality of optimum division points of the character image and an optimum division path for each optimum division point from the candidate division points and candidate division paths at the candidate division points,
Searching for an optimum division point and an optimum division path for each optimum division point from the candidate division points and the candidate division points at the candidate division points by dynamic programming based on the two-dimensional label for each candidate division path. Item 3. A character recognition method according to Item 2. Recognizing the one or more image segments and obtaining one or more recognition results,
Classifying the one or more image segments;
The character recognition method according to claim 1, further comprising: calculating a matching distance between each type of image segment and a known character image of that type for each image segment as a recognition result of the image segment. The character recognition device
Before performing the step of recognizing the one or more image segments and obtaining one or more recognition results, merge adjacent image segments and / or one of the plurality of image segments. The character recognition method according to claim 1, further comprising executing a step of performing division again on one or more. A candidate path specifying unit for specifying one or a plurality of candidate division paths for each candidate division point in the character image to be recognized;
An image segment generation unit that divides the character image to be recognized by each candidate division pass and generates one or a plurality of image segments;
A recognition unit that recognizes each of the one or more image segments and obtains one or more recognition results;
A path search unit that selects one or a plurality of optimum division points of the character image and an optimum division path for each optimum division point from the candidate division points and candidate division paths at the candidate division points based on the recognition result; A character recognition device comprising: The candidate path specifying unit further includes:
One or more candidate divisions for each of the candidate division points using a two-dimensional marker that includes a first indicator for marking each candidate division point and a second indicator for marking each candidate division path The character recognition device according to claim 6, which is arranged so as to mark each of the paths. The path search unit further includes:
The character recognition device according to claim 7, wherein the character recognition device is arranged so as to search for an optimum division point and an optimum division path from the candidate division points and the candidate division paths by dynamic programming based on the two-dimensional mark. The recognition unit
A classifier for classifying the one or more image segments;
The character recognition of Claim 6 provided with the matching part which calculates the matching distance between the known character image of the said kind for every image segment based on the kind of each said image segment as a recognition result of the segment of the said image apparatus. The image segment generation unit further includes
Arranging adjacent image segments and / or performing re-division on one or more of the plurality of image segments and outputting the resulting image segments to the character recognition device The character recognition device according to claim 6.

Images