JP5867227B2 - Learning data generation device for character recognition, character recognition device, and program - Google Patents

Description

  The present invention relates to a character recognition learning data generation device, a character recognition device, and a program.

  Patent Document 1 describes a Pricing style in which characters entered by online handwritten character input are written separately, a Cursive style in which characters are continuously written, a Pricing style and a Cursive style. A handwritten character recognizing device that recognizes separately is disclosed. Patent Document 2 discloses a handwriting recognition device that improves the recognition rate and the recognition processing speed for continuously broken characters by using a dictionary in which a pattern that is statistically easy to be drawn is registered. In Patent Document 3, in recognition processing of handwritten characters input from a touch panel or the like, recognition processing using a threshold value for distinguishing between uppercase letters such as “i” and “ro” and lowercase letters such as “tsu” is performed. A handwritten character recognition device that improves the recognition rate of characters of different sizes has been disclosed. Patent Document 4 discloses an adjustment method for adjusting a line space and a base line in an alphabet recognition process such as “h”, “g”, etc. whose width varies vertically.

JP-A-11-167606 JP-A-8-249424 JP-A-9-231316 Japanese Patent Laid-Open No. 7-6204

  One of the objects of the present invention is to provide a character recognition learning data generation device, a character recognition device, and a character recognition device that reduce the influence of a difference in character size when performing character recognition of a document containing characters of different sizes. To provide a program.

  In order to achieve the above object, the invention described in claim 1 includes a plurality of character elements having a plurality of character elements, the order of which is defined, and the plurality of learning objects composed of the plurality of character elements. Character element information indicating the position and size of each of the character elements in the learning target document, and corresponding character information indicating which of the plurality of character elements corresponds to each of the plurality of characters. Character element information acquisition means, for the two character elements that are consecutive in the order among the plurality of character elements, vector information indicating a vector from the preceding character element to the subsequent character element in the order in the character element information Vector information generating means for generating the information based on the character element information, the size of at least one of the two consecutive character elements indicated by the character element information The vector information correcting means for correcting the size of the vector and the determination target document to be determined determine whether or not two character elements included in the determination target document belong to the same character. Means for generating learning data to be used at the time, information indicating whether or not the two consecutive character elements specified by the corresponding character information are included in the same character, and correction by the vector information correction unit A learning data generation device for character recognition, comprising learning means for generating the learning data using input vector information as input data.

  The invention according to claim 2 is the learning data generation apparatus for character recognition according to claim 1, wherein the vector information correction unit circumscribes at least one of the two consecutive character elements. The character recognition learning data generating apparatus is characterized in that the size of the vector is corrected in accordance with a ratio of the shape to a predetermined shape.

  The invention according to claim 3 is the learning data generation device for character recognition according to claim 1 or 2, wherein the character element information acquisition unit is formed first for each of the character elements. Further, end position information indicating the position of the part and the position of the last formed part is further acquired, and the vector information generation unit is formed at the end of the preceding character element based on the end part position information. A character recognition learning data generating apparatus that generates vector information indicating a vector from a part to a first formed part of the succeeding character element.

  The invention according to claim 4 is the learning data generation apparatus for character recognition according to claim 1 or 2, wherein the vector information generation means is a center point of a rectangular area circumscribing the preceding character element. The learning data generation device for character recognition is characterized by generating vector information indicating a vector that reaches a center point of a rectangular area circumscribing the subsequent character element.

  In addition, the invention according to claim 5 includes a plurality of character elements in which an order is defined, and a determination target document including a plurality of characters constituted by the plurality of character elements. Character element information acquisition means for acquiring character element information indicating a position and a size in a document to be determined, among two character elements that are consecutive in the order among the plurality of character elements, from character elements that precede in the order Vector information generating means for generating vector information indicating a vector toward the subsequent character element based on the character element information, according to the size of at least one of the two consecutive character elements indicated by the character element information Corrected by the vector information correcting means for correcting the magnitude of the vector and the vector information correcting means. Based on the vector information, determining means for determining whether the two character elements belong to the same character, a character recognition apparatus characterized by having a.

  In the invention according to claim 6, the computer includes a plurality of character elements with respect to a learning object document including a plurality of characters including a plurality of character elements, the order of which is defined. Character element information for acquiring character element information indicating the position and size in each of the learning target documents and corresponding character information indicating which of the plurality of characters corresponds to each of the plurality of character elements An acquisition unit generates, based on the character element information, vector information indicating a vector from the preceding character element to the succeeding character element in the order for two character elements that are consecutive in the order among the plurality of character elements. Vector information generating means for performing at least one size of the two consecutive character elements indicated by the character element information Next, when determining whether or not two character elements included in the determination target document belong to the same character for the vector information correction means for correcting the size of the vector and the determination target document to be determined Means for generating learning data for use in the information, which is specified by the corresponding character information and indicating whether or not the two consecutive character elements are included in the same character, and is corrected by the vector information correcting unit. This is a program for functioning as learning means for generating the learning data using the vector information as input data.

  In the invention according to claim 7, the computer includes a plurality of character elements with respect to a determination target document including a plurality of character elements including a plurality of character elements, the order of which is defined. Character element information acquisition means for acquiring character element information indicating a position and a size in each document to be judged, and preceding two character elements in the order among the plurality of character elements. Vector information generating means for generating vector information indicating a vector from a character element to a subsequent character element based on the character element information, the size of at least one of the two consecutive character elements indicated by the character element information And a vector information correcting means for correcting the magnitude of the vector according to the size, and the vector information correcting means. Based on the corrected vector information by the two character elements are programmed for causing the determining means for determining whether or not belonging to the same character function as.

  According to the first and sixth aspects of the invention, the learning data used when determining whether or not two character elements are included in the same character uses vector information corrected according to the size of the character element. Generated.

  According to the invention of claim 2, the size of the vector is corrected according to the size of at least one of the character element having the start point and the end point of the vector.

  According to the invention of claim 3, vector information indicating a vector from the last formed part of the preceding character element to the first formed part of the subsequent character element among the two consecutive character elements. Is generated.

  According to the fourth aspect of the present invention, the vector from the center point of the rectangular area circumscribing the preceding character element to the center point of the rectangular area circumscribing the succeeding character element is shown among the two consecutive character elements. Vector information is generated.

  According to the fifth and seventh aspects of the present invention, the determination as to whether or not two character elements are included in the same character is performed using vector information corrected according to the size of the character element.

It is a figure which shows the structure of the learning data generation apparatus for character recognition which concerns on the 1st Embodiment of this invention. It is a figure which shows the example of the stroke shown by stroke information. It is a figure which shows an example of correction | amendment of an off stroke. It is a flowchart which shows operation | movement of the learning data generation apparatus for character recognition which concerns on the 1st Embodiment of this invention. It is a figure which shows the structure of the character recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of the character recognition apparatus which concerns on the 2nd Embodiment of this invention. It is a figure which shows the example of the character element extracted in an offline process.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[First Embodiment]
FIG. 1 is a diagram showing a configuration of a character recognition learning data generation apparatus 100 according to the first embodiment of the present invention. The character recognition learning data generation apparatus 100 is connected to an input reception apparatus 200 that recognizes a user's handwriting action such as a tablet PC or an electronic pen.

  The learning data generation device for character recognition 100 includes an external I / F unit 110, a stroke information acquisition unit 120, an off-stroke information generation unit 130, an off-stroke information correction unit 140, a learning processing unit 150, and a storage unit 160. The external I / F unit 110, the stroke information acquisition unit 120, the off-stroke information generation unit 130, the off-stroke information correction unit 140, and the learning processing unit 150 are functions of a CPU that operates according to a program stored in the storage unit 160. Realized. The storage unit 160 includes a storage device such as a hard disk or a memory.

  In the input receiving device 200, a writing tool (touch pen, electronic pen, fingertip, etc.) is lowered to a recording medium (tablet PC display, paper, etc.) and released from (pen down, start of contact) (end of pen up, end of contact). The electric signal indicating the movement of the tip of the writing tool up to) is generated and output to the learning data generation device 100 for character recognition. Further, for example, when a button provided on the writing tool is pressed between writing of one character and writing of the next character, a signal indicating the switching of characters is generated, and the character recognition learning data generating device 100 is notified. Output.

  Here, what is input by the writing tool is written document data to be learned consisting of hundreds to thousands of characters given in advance.

  The external I / F unit 110 acquires a signal input from the input reception device 200 and outputs the signal to the stroke information acquisition unit 120.

  The stroke information acquisition unit 120 corresponds to any of a plurality of characters in the learning target document, and stroke (character element) information indicating the position and size of each of the plurality of character elements in the learning target document. It operates as a character element information acquisition means for acquiring the corresponding character information indicating. In other words, the stroke information acquisition unit 120 is based on a signal input from the external I / F unit 110, and the strokes constituting each character included in the learning target document data (between pen down and pen up (one stroke)). Stroke information indicating the position, shape, and size of the recorded character element) in the learning target document is acquired. Among these, the information indicating the position and shape is acquired as coordinate data of the position where the writing instrument comes into contact with the recording medium. (The shape is acquired as coordinate data acquired at a predetermined cycle when moving while in contact.) In addition, the information indicating the position and shape includes the first part of the stroke (pen down was performed) Part) and end part information indicating the positions of the last part (part where the pen-up has been performed) are included. For the information indicating the size, the difference between the maximum value and minimum value in the horizontal direction (x-axis direction) and the maximum value and minimum value in the vertical direction (y-axis direction) of all coordinates acquired during one stroke. Obtained as a value difference. Further, character correspondence information indicating which character (what number of character) each stroke corresponds to is acquired from the signal based on the operation at the time of character switching described above. Thus, stroke information is continuously input based on the operation of the writing instrument, and the order is defined in a plurality of strokes by performing acquisition and recording in accordance with the input order.

  FIG. 2 is a diagram illustrating an example of a stroke indicated by the stroke information. In FIG. 2, strokes 301, 302, and 303 correspond to the first character “A”, and strokes 304 and 305 correspond to the second character “I”, respectively. The order is defined in the order of strokes 301, 302, 303, 304, and 305.

  The off-stroke information generation unit 130 converts, for the stroke information, off-stroke information indicating a vector (off-stroke) from the preceding stroke to the succeeding stroke in the order with respect to two strokes (character elements) that are consecutive in the above-described order. Generate based on. The off-stroke is the point at which the writing instrument records a stroke and pen-up (the part formed at the end of the stroke), then the pen-down point (the part formed at the beginning of the next stroke) It is a vector that reaches the position of. The off-stroke information can be obtained by subtracting the coordinates of the last point of the stroke from the coordinates of the first point of the next stroke. In strokes 301, 302, 303, 304, and 305, off-strokes 401, 402, 403, 404, and 405 are obtained.

  The off-stroke information correcting unit 140 indicates the off-stroke indicated by the off-stroke information generated by the off-stroke information generating unit 130 according to the magnitude of at least one of the two consecutive strokes indicated by the stroke information. It operates as a vector information correcting means for correcting the magnitude of (vector). The off-stroke information correction unit 140 corrects the size of each of the plurality of off-strokes indicated by the off-stroke information according to the size of the stroke having the start point of the off-stroke. Specifically, a rectangular area circumscribing the corresponding stroke is converted into a predetermined shape such as a square having a vertical and horizontal length of 1 unit (unit may be appropriately defined), for example. , Transform off-stroke. FIG. 3 is a diagram illustrating an example of off-stroke correction. For example, in FIG. 2, when the size of the stroke 303 (that is, the size of the rectangular area circumscribing the stroke 303) is 2 units in the vertical direction and 4 units in the horizontal direction, Deform to 1/4 in the direction. Further, the off-stroke information correction unit 140 also corrects the stroke information by deforming the stroke indicated by the stroke information (that is, the vertical and horizontal lengths are set as one unit). As a result of this correction, for example, when the size of the stroke is greatly different from each other due to different character sizes in the document to be learned, the sizes of the stroke and the off-stroke are made uniform. A process equivalent to the process for a document to be learned consisting of characters of a large size is executed.

  The learning processing unit 150 operates as a learning unit that generates learning data used when determining whether or not two strokes included in the determination target document belong to the same character. The learning processing unit 150 acquires stroke information and off-stroke information corrected by the off-stroke information correcting unit 140, respectively. Based on the character correspondence information included in the acquired stroke information, the learning processing unit 150 determines whether the off-stroke is a transition between characters (transition between characters), a transition within the same character (transition within a character), Determine.

  The learning processing unit 150 associates the corrected off-stroke information related to the off-stroke determined to be the transition between characters and the corrected stroke information related to the stroke having the start point of the off-stroke into the “positive data list”, The corrected off-stroke information related to the off-stroke determined to be the internal transition and the corrected stroke information related to the stroke having the start point of the off-stroke are associated with each other and added to the “negative data list”. The learning processing unit 150 uses these pieces of information to determine whether the off-stroke is an off-stroke in a character or a character in a known character recognition process such as SVM (support vector machine) which is an identification method used for supervised learning. Learning data to be used in determining whether the stroke is between is generated. The generated learning data is stored in the storage unit 160.

  Up to this point, the learning processing unit 150 classifies the corrected off-stroke information related to a certain off-stroke (target off-stroke) and the corrected stroke information related to the stroke having the start point of the off-stroke, and uses them in the learning process. Although the configuration is shown, the learning processing unit 150 has corrected the corrected off-stroke information regarding the target off-stroke and the predetermined number of strokes (n: n is an integer of 1 or more) preceding the target off-stroke. If the stroke information and the corrected stroke information related to a predetermined number of strokes (m: m is an integer equal to or greater than 1) following the target stroke are classified in association with each other and used in the learning process, the number is more. A learning process based on the data is executed.

  Here, the corrected off-stroke information regarding the target off-stroke, the corrected stroke information regarding the n strokes preceding the target off-stroke, and the corrected stroke regarding the m strokes following the target stroke. The operation of the learning data generating apparatus for character recognition 100 that is classified in association with information and used for learning will be described with reference to a flowchart. FIG. 4 is a flowchart showing the operation of the character recognition learning data generation apparatus 100 according to the first embodiment of the present invention.

First, the stroke information acquisition unit 120 acquires stroke information based on a signal input from the input receiving apparatus 200 and acquired by the external I / F unit 110 (S401). Here, the number of strokes for which stroke information is acquired is N (N: an integer equal to or greater than 2), and the stroke is represented by St _i (0 ≦ i <N).

Next, the off-stroke information generating unit 130 generates off-stroke information based on the stroke information acquired in S401 (S402). Here, N-1 pieces of off-stroke information are generated, and hereinafter, these off-strokes are represented by OSt _j (0 ≦ j <N−1). Then, the off-stroke information correction unit 140 corrects the stroke information acquired in S401 and the off-stroke information generated in S402 (S403).

Then, the learning processing unit 150 applies corrected strokes St _j−n , St _{j−n + 1} ,..., St _j ,..., St _{j + m−1} , St _{j + m} for each of the off strokes OSt _j generated in S402. The information shown is acquired (S404). Next, the learning processing unit 150 determines whether the vector OSt _j is an inter-character transition or an intra-character transition (S405). If it is an inter-character transition, each information is classified into a positive data list. (S406) If it is an intra-character transition, each information is classified into a negative data list (S407). The learning processing unit 150 executes the processing from S404 to S407 for all OSt _j for which the offstroke information was acquired in S402, and then the learning processing unit 150 executes the learning processing (S408), and the generated learning data is used. It preserve | saves at the memory | storage part 160 (S409), and the operation | movement of the learning data generation apparatus 100 for character recognition is complete | finished.

[Second Embodiment]
FIG. 5 is a diagram showing a configuration of a character recognition device 500 according to the second embodiment of the present invention. The character recognition device 500 is connected to an input reception device 600 that recognizes a user's handwriting action such as a tablet PC or an electronic pen.

  The character recognition device 500 includes an external I / F unit 510, a stroke information acquisition unit 520, an off-stroke information generation unit 530, an off-stroke information correction unit 540, a recognition processing unit 550, and a storage unit 560. The external I / F unit 510, the stroke information acquisition unit 520, the off-stroke information generation unit 530, the off-stroke information correction unit 540, and the recognition processing unit 550 are functions of a CPU that is operated by a program stored in the storage unit 560. Realized. The storage unit 560 includes a storage device such as a hard disk or a memory.

  The input reception device 600 generates an electrical signal indicating the movement of the tip of the writing tool from when the writing tool is lowered to the recording medium until it is released, and outputs the electrical signal to the character recognition device 500. Further, for example, when a button provided on the writing tool is pressed between writing of a certain character and writing of the next character, a signal indicating character switching is generated and output to the character recognition device 500.

  Here, what is input by the writing implement is determination target document data that is an object of character recognition.

  External I / F unit 510 acquires a signal input from input reception device 600 and outputs the signal to stroke information acquisition unit 520.

  The stroke information acquisition unit 520 corresponds to any of a plurality of characters in the learning target document, and stroke (character element) information indicating the position and size of each of the plurality of character elements in the learning target document. It operates as a character element information acquisition means for acquiring the corresponding character information indicating. That is, the stroke information acquisition unit 520 is based on the signal input from the external I / F unit 510, and the strokes constituting each character included in the determination target document data (between pen down and pen up (one stroke)). Stroke information indicating the position, shape, and size of the recorded character element) in the determination target document is acquired. Among these, the information indicating the position and shape is acquired as coordinate data of the position where the writing instrument comes into contact with the recording medium. (The shape is acquired as coordinate data acquired at a predetermined cycle when moving while in contact.) In addition, the information indicating the position and shape includes the first part of the stroke (pen down was performed) Part) and end part information indicating the positions of the last part (part where the pen-up has been performed) are included. For the information indicating the size, the difference between the maximum value and minimum value in the horizontal direction (x-axis direction) and the maximum value and minimum value in the vertical direction (y-axis direction) of all coordinates acquired during one stroke. Obtained as a value difference. Further, character correspondence information indicating which character (what number of character) each stroke corresponds to is acquired from the signal based on the operation at the time of character switching described above. Thus, stroke information is continuously input based on the operation of the writing instrument, and the order is defined in a plurality of strokes by performing acquisition and recording in accordance with the input order.

  The off-stroke information generation unit 530 uses, as stroke information, off-stroke information indicating a vector (off-stroke) from the preceding stroke to the succeeding stroke in the order for two consecutive strokes (character elements) in the above-described order. Generate based on. The off-stroke is the point at which the writing instrument records a stroke and pen-up (the part formed at the end of the stroke), then the pen-down point (the part formed at the beginning of the next stroke) It is a vector that reaches the position of. The off-stroke information can be obtained by subtracting the coordinates of the last point of the stroke from the coordinates of the first point of the next stroke.

  The off-stroke information correcting unit 540 indicates the off-stroke indicated by the off-stroke information generated by the off-stroke information generating unit 130 according to the magnitude of at least one of the two consecutive strokes indicated by the stroke information. It operates as a vector information correcting means for correcting the magnitude of (vector). The off-stroke information correction unit 540 corrects the size of each of the plurality of off-strokes indicated by the off-stroke information according to the size of the stroke having the start point of the off-stroke. Specifically, a rectangular area circumscribing the corresponding stroke is converted into a predetermined shape such as a square having a vertical and horizontal length of 1 unit (unit may be appropriately defined), for example. , Transform off-stroke. The off-stroke information correction unit 540 also corrects the stroke information by deforming the stroke indicated by the stroke information (that is, taking the vertical and horizontal lengths as one unit). As a result of this correction, for example, when the size of the stroke is greatly different from each other due to different character sizes in the document to be learned, the sizes of the stroke and the off-stroke are made uniform. A process equivalent to the process for a document to be learned consisting of characters of a large size is executed.

  The recognition processing unit 550 operates as a determination unit that determines whether two strokes included in a recognition target document belong to the same character. The recognition processing unit 550 acquires stroke information and off-stroke information corrected by the off-stroke information correcting unit 540, respectively. Based on the acquired stroke information and off-stroke information, the recognition processing unit 550 determines whether the off-stroke is a transition within the same character or a transition between characters. Specifically, recognition by a known recognition method such as SVM (support vector machine) is performed for corrected off-stroke information of a certain off-stroke and corrected stroke information of a stroke having the start point of the off-stroke. A process is performed to determine whether the off-stroke is a transition within the same character or a transition between characters.

  Then, the recognition processing unit 550 determines that the characters are switched in the off-stroke determined to be a transition between characters, and a single character that is converted into a text code by a known technique with respect to the stroke group delimited by the transition between the characters. Perform recognition processing. Furthermore, when a plurality of recognition result (character) candidates are generated for each character to be determined by single character recognition processing, context processing is performed to select and correct the recognition results in consideration of the probability of language To do. The context processing includes a character string list, an n-gram list, a regular expression, and the like.

  Up to this point, a configuration has been shown in which the recognition processing unit 550 performs determination processing using corrected off-stroke information regarding a certain off-stroke (target off-stroke) and corrected stroke information regarding a stroke having the start point of the off-stroke. However, the recognition processing unit 550 includes the corrected off-stroke information related to the target off-stroke, the corrected stroke information related to a predetermined number of strokes (n: n is an integer of 1 or more) preceding the target off-stroke, and the If the determination process is performed using corrected stroke information related to a predetermined number of strokes (m: m is an integer equal to or greater than 1) following the target stroke, the determination process based on more data is performed. Executed.

  Here, the recognition processing unit 550 corrects the corrected off-stroke information related to the target off-stroke, the corrected stroke information related to n strokes preceding the target off-stroke, and the corrected stroke related to m strokes subsequent to the target stroke. The operation of the character recognition device 500 that executes learning processing using information will be described using a flowchart. FIG. 6 is a flowchart showing the operation of the character recognition apparatus 500 according to the second embodiment of the present invention.

First, the stroke information acquisition unit 520 acquires stroke information based on a signal input from the input receiving device 600 and acquired by the external I / F unit 510 (S601). Here, the number of strokes for which stroke information is acquired is N (N: an integer equal to or greater than 2), and the stroke is represented by St _i (0 ≦ i <N).

Next, the off-stroke information generation unit 530 generates off-stroke information based on the stroke information acquired in S601 (S602). Here, N-1 pieces of off-stroke information are generated, and hereinafter, these off-strokes are represented by OSt _j (0 ≦ j <N−1). Then, the off-stroke information correction unit 540 corrects the stroke information generated in S602 (S603).

Then, for each of the off strokes OSt _j generated in S602, the recognition processing unit 550 indicates information indicating corrected strokes St _i−n , St _{i−n + 1} ,..., St _i ,..., St _{i + m−1} , St _{i + m.} Is acquired (S604). The recognition processing unit 550 determines whether the off-stroke OSt _i is an intra-character transition or an inter-character transition (S605). The recognition processing unit 550 executes the processing of S604 and S605 for all OSt _j for which the off stroke information has been acquired in S602.

Next, the recognition processing unit 550 executes the character recognition process and the context process (S606) for each of the stroke groups delimited by the off-stroke OSt _j determined as the inter-character transition in S605, so that the character recognition device 500 The operation ends.

  With the above configuration, learning data generation processing and character recognition processing that reduce the influence of the difference in character size are executed when character recognition is performed on a document that includes characters of different sizes.

  In addition, said embodiment demonstrates the principle of this invention, an effect, and a function as an example, and this invention is not limited by these. For example, in the above embodiment, stroke information indicating the shape of each stroke is acquired by online processing using a writing instrument such as a touch pen, and a vector from the end point to the start point of these strokes is used as an off-stroke. Although shown, it is good also as a structure which performs the production | generation process and the character recognition process of learning data by the offline process using the document already described. The configuration in that case is shown below.

  In the case of offline processing, the stroke information acquisition units 120 and 520 scan the written document, perform binarization, and perform character element extraction processing. In the character element extraction processing, for example, each extracted by performing straight line or curve extraction processing may be used as a character element, or a block of drawn dots connected to each other may be used as one character element. FIG. 7 is a diagram illustrating an example of character elements extracted in the offline processing. In FIG. 7, character elements 701, 702, and 703 are extracted. The off-stroke information generation units 130 and 530 define the order of the extracted character elements 701, 702, and 703 in accordance with predetermined rules such as top to bottom and left to right (here, the character element 701). The character elements 702 and the character elements 703 are in this order.) According to this order, vectors connecting the centers of the character elements 701, 702, and 703 are generated as off-strokes 801 and 802. Thereafter, correction processing and learning data generation processing or character recognition processing are executed as in the first and second embodiments described above.

  In the embodiment described so far, the configuration in which the off-stroke correction is performed based on the size of the stroke (character element) having the start point of the off-stroke is disclosed. However, the stroke having the end point of the off-stroke is disclosed. It is good also as a structure performed based on the magnitude | size of this, and it is good also as a structure performed based on the magnitude | size of both the stroke which has the starting point of an offstroke, and the stroke which has the end point of an offstroke.

  The operations of the character recognition learning data generation device 100 and the character recognition device 500 described here are the programs stored in the storage units 160 and 560 of the character recognition learning data generation device 100 and the character recognition device 500, respectively. Realized by operating. This program may be provided by communication, or may be provided by being stored in a storage medium such as a CD-ROM that can be read by a computer.

  100 character recognition learning data generation device, 110 external I / F unit, 120 stroke information acquisition unit, 130 off stroke information generation unit, 140 off stroke information correction unit, 150 learning processing unit, 160 storage unit, 200 input reception device, 301, 302, 303, 304, 305 Stroke, 401, 402, 403, 404, 405 Off stroke, 500 character recognition device, 510 External I / F unit, 520 Stroke information acquisition unit, 530 Off stroke information generation unit, 540 Off Stroke information correction unit, 550 recognition processing unit, 560 storage unit, 600 input reception device, 701, 702, 703 character element, 801, 802 off-stroke.

Claims (7)

A learning target document including a plurality of character elements including a plurality of character elements, the order of which is defined, and indicating the position and size of each of the plurality of character elements in the learning target document. Character element information acquisition means for acquiring character element information and corresponding character information indicating which of the plurality of characters corresponds to each of the plurality of character elements;
Vector information for generating vector information indicating a vector from the preceding character element to the subsequent character element in the order, based on the character element information, for two character elements that are consecutive in the order among the plurality of character elements Generating means,
Vector information correction means for correcting the size of the vector in accordance with the size of at least one of the two consecutive character elements indicated by the character element information; and
A means for generating learning data used for determining whether or not two character elements included in the determination target document belong to the same character with respect to the determination target document to be determined, specified by the corresponding character information Learning that generates the learning data using, as input data, information indicating whether or not the two consecutive character elements are included in the same character and the vector information corrected by the vector information correcting unit means,
A character recognition learning data generation device characterized by comprising: The learning data generating device for character recognition according to claim 1,
The vector information correcting unit corrects the size of the vector according to a ratio of a shape of a rectangular area circumscribing at least one of the two consecutive character elements to a predetermined shape. A learning data generator for recognition. The learning data generation device for character recognition according to claim 1 or 2,
The character element information acquisition means further acquires end position information indicating the position of the first formed portion and the position of the last formed portion of each of the character elements,
The vector information generation means, based on the end position information, vector information indicating a vector from the last formed part of the preceding character element to the first formed part of the subsequent character element. A learning data generating device for character recognition characterized by generating. The learning data generation device for character recognition according to claim 1 or 2,
The vector information generating means generates vector information indicating a vector from a central point of a rectangular area circumscribing the preceding character element to a central point of the rectangular area circumscribing the subsequent character element. Learning data generation device for character recognition. A determination target document including a plurality of character elements including a plurality of character elements, the order of which is defined, and indicating the position and size of each of the plurality of character elements in the determination target document. Character element information acquisition means for acquiring character element information;
Vector information for generating vector information indicating a vector from the preceding character element to the subsequent character element in the order, based on the character element information, for two character elements that are consecutive in the order among the plurality of character elements Generating means,
Vector information correction means for correcting the size of the vector in accordance with the size of at least one of the two consecutive character elements indicated by the character element information; and
Determining means for determining whether or not the two character elements belong to the same character based on the vector information corrected by the vector information correcting means;
A character recognition device comprising: Computer
A learning target document including a plurality of character elements including a plurality of character elements, the order of which is defined, and indicating the position and size of each of the plurality of character elements in the learning target document. Character element information acquisition means for acquiring character element information and corresponding character information indicating which of the plurality of characters corresponds to each of the plurality of character elements;
Vector information for generating vector information indicating a vector from the preceding character element to the subsequent character element in the order, based on the character element information, for two character elements that are consecutive in the order among the plurality of character elements Generating means,
Vector information correction means for correcting the size of the vector in accordance with the size of at least one of the two consecutive character elements indicated by the character element information; and
A means for generating learning data used for determining whether or not two character elements included in the determination target document belong to the same character with respect to the determination target document to be determined, specified by the corresponding character information Learning that generates the learning data using, as input data, information indicating whether or not the two consecutive character elements are included in the same character and the vector information corrected by the vector information correcting unit means,
Program to function as. Computer
A determination target document including a plurality of character elements including a plurality of character elements, the order of which is defined, and indicating the position and size of each of the plurality of character elements in the determination target document. Character element information acquisition means for acquiring character element information;
Vector information for generating vector information indicating a vector from the preceding character element to the subsequent character element in the order, based on the character element information, for two character elements that are consecutive in the order among the plurality of character elements Generating means,
Vector information correction means for correcting the size of the vector in accordance with the size of at least one of the two consecutive character elements indicated by the character element information; and
Determining means for determining whether or not the two character elements belong to the same character based on the vector information corrected by the vector information correcting means;
Program to function as.

Images