JP5471254B2 - Verification device, verification method, verification program, and creation device - Google Patents

Description

  The present invention relates to a verification device, a verification method, a verification program, and a creation device.

  2. Description of the Related Art Conventionally, there is a character recognition device that optically acquires a document in which characters are described as image data and recognizes characters included in the acquired image data. There is also a verification device that verifies whether a character obtained as a result of character recognition is correct using a distance value after character recognition processing by the character recognition device. Hereinafter, the image data is referred to as a character image.

  Here, the verification apparatus will be further described. The verification device has a dictionary in which feature amounts for each character are registered. Then, the verification device calculates a distance value indicating a distance between the feature amount calculated for the character included in the character image and the feature amount previously registered in the dictionary for the character obtained as a character recognition result. The verification device verifies that the smaller the distance value is, the higher the probability of correctness is compared to the case where the distance value is large. Note that the feature amount indicates a parameter relating to the shape of the character, such as the inclination of the character line and the character area.

Japanese Patent Laid-Open No. 10-63784

  However, the verification device described above has a problem that the accuracy of the verification result is poor. In particular, when verifying similar characters that have other similar characters, the accuracy of the verification results is poor. The similar characters correspond to, for example, “l (el)” and “1 (1)” or “i” and “i”.

  For example, a case where the character included in the character image is “l” and the character obtained as a character recognition result is “1” will be described as an example. Here, “l (el)” and “1 (1)” are similar in shape, and the feature amount calculated for “l (el)” and “1 (1)” are registered in the dictionary in advance. The distance value between the set feature amount may be small. In this case, the verification device described above verifies that the probability of correctness is high. That is, the verification device described above may verify that the probability of correctness is high even if the character obtained as a result of character recognition is incorrect.

  The disclosed technology has been made in view of the above, and an object thereof is to provide a verification device, a verification method, a verification program, and a creation device capable of accurately verifying whether a character obtained as a character recognition result is correct. And

  In one aspect, the disclosed verification apparatus includes a character recognition unit that executes character recognition processing on an input character image when the character image is input. In addition, the verification device includes a condition for distinguishing the first character from the second character that may be obtained as a result of erroneous recognition in the character recognition process for the first character, the first character, and the first character For each character of the second character, information indicating the size of the character included in the character image in the character image, information indicating the relationship between the character and other nearby characters, and the character When the result of the character recognition process of the character included in the character image is the first character using the attribute value including at least one of the information indicating the probability of the result of the character recognition process A creation unit for creating a verification expression for verifying the correctness of the result is provided. Further, the verification device identifies whether the first character is included in the result of the character recognition processing by the character recognition unit, and when the verification device identifies that the first character is included, the verification formula created by the creation unit The verification part which verifies using is provided.

  According to one aspect of the disclosed verification apparatus, there is an effect that it is possible to accurately verify whether a character obtained as a character recognition result is correct.

FIG. 1 is a block diagram illustrating an example of the configuration of the verification apparatus according to the first embodiment. FIG. 2 is a block diagram illustrating an example of the configuration of the verification apparatus according to the second embodiment. FIG. 3 is a diagram illustrating an example of information stored in the learning data table according to the second embodiment. FIG. 4 is a schematic diagram illustrating an example of information stored in the best integrated logical expression table according to the second embodiment. FIG. 5 is a diagram illustrating an example of a display screen according to the second embodiment. FIG. 6A is a schematic diagram illustrating an overall image of processing performed by the best logical expression creation unit according to the second embodiment. FIG. 6B is a diagram for explaining the overall image of the process performed by the best logical expression creation unit according to the second embodiment. FIG. 6C is a schematic diagram illustrating an overall process performed by the best logical expression creating unit according to the second embodiment. FIG. 6-4 is a schematic diagram illustrating an overall process performed by the best logical expression creating unit according to the second embodiment. FIG. 7A is a schematic diagram illustrating generalization processing according to the second embodiment. FIG. 7B is a schematic diagram illustrating generalization processing according to the second embodiment. FIG. 8 is a diagram illustrating processing for selecting a logical expression having the highest evaluation value in the second embodiment. FIG. 9A is a schematic diagram illustrating an overall process performed by the best integrated logical expression creation unit according to the second embodiment. FIG. 9-2 is a schematic diagram illustrating an overall process performed by the best integrated logical expression creation unit according to the second embodiment. FIG. 9C is a schematic diagram illustrating an overall process performed by the best integrated logical expression creation unit according to the second embodiment. FIG. 9-4 is a schematic diagram illustrating an overall process performed by the best integrated logical expression creating unit according to the second embodiment. FIG. 10 is a flowchart illustrating an example of the flow of best integrated logical expression creation processing according to the second embodiment. FIG. 11 is a flowchart illustrating an example of the flow of the best logical expression creation process in the second embodiment. FIG. 12 is a flowchart for explaining an example of the flow of generalization processing in the second embodiment. FIG. 13 is a flowchart illustrating an example of a processing flow for selecting a logical expression having the highest evaluation value in the second embodiment. FIG. 14 is a flowchart illustrating an example of the flow of verification processing according to the second embodiment. FIG. 15 is a schematic diagram illustrating an example of a computer that executes a verification program according to the second embodiment.

  Hereinafter, embodiments of the disclosed verification apparatus, verification method, verification program, and creation apparatus will be described in detail with reference to the drawings. Note that the invention disclosed by this embodiment is not limited. Each embodiment can be appropriately combined within a range in which processing contents do not contradict each other.

  An example of the configuration of the verification apparatus 100 according to the first embodiment will be described with reference to FIG. FIG. 1 is a block diagram illustrating an example of the configuration of the verification apparatus according to the first embodiment. In the example illustrated in FIG. 1, the verification apparatus 100 includes a creation unit 101, a character recognition unit 102, and a verification unit 103.

  The creation unit 101 uses a condition for distinguishing the second character and the first character that may be obtained as a result of erroneous recognition in the character recognition process for the first character, and the character image using the attribute value. If the result of the character recognition process for the character included in the first character is a verification formula for verifying the correctness of the result. Here, the attribute value includes, for each character of the first character and the second character, information indicating the size of the character included in the character image in the character image and other characters in the vicinity of the character. And at least one of information indicating the likelihood of the result of character recognition processing for the character.

  When a character image is input, the character recognition unit 102 performs a character recognition process on the input character image. The verification unit 103 identifies whether or not the first character is included in the result of the character recognition processing by the character recognition unit, and when the verification unit 103 identifies that the first character is included, the verification unit 103 uses the verification formula created by the creation unit 101 Perform verification.

  That is, when the character recognition result includes a character that is easily misrecognized, the verification apparatus 100 uses the verification formula created in advance by taking various information into account, and performs the misrecognition obtained as the character recognition result. Verify that easy-to-read characters are correct as character recognition results. As a result, according to the verification apparatus 100 according to the first embodiment, it is possible to accurately verify whether the character obtained as the character recognition result is correct.

[Configuration of Verification Device According to Second Embodiment]
A verification apparatus 200 according to the second embodiment will be described. First, an example of the configuration of the verification apparatus 200 according to the second embodiment will be described with reference to FIG. FIG. 2 is a block diagram illustrating an example of the configuration of the verification apparatus according to the second embodiment. In the example illustrated in FIG. 2, the verification device 200 includes an input unit 201, a display unit 202, a storage unit 300, and a control unit 400.

  The input unit 201 is connected to the control unit 400. The input unit 201 receives input of information from the user and sends the received information to the control unit 400. The input unit 201 corresponds to a keyboard, mouse, microphone, or an image scanner or camera that acquires a character image of a document. Display unit 202 is connected to control unit 400. The display unit 202 receives information from the control unit 400 and displays the received information to the user. The display unit 202 corresponds to a monitor (or display, touch panel) or the like.

  The details of the information received by the input unit 201 and the details of the information displayed by the display unit 202 will be omitted here, and will be described together with the description of each related unit.

  The storage unit 300 is connected to the control unit 400 and stores data used for various processes performed by the control unit 400. The storage unit 300 is, for example, a semiconductor memory device such as a random access memory (RAM), a read only memory (ROM), or a flash memory, or a storage device such as a hard disk or an optical disk. In the example illustrated in FIG. 2, the storage unit 300 includes a learning data table 301 and a best integrated logical expression table 302.

  The learning data table 301 stores in advance attribute values related to characters included in the character image in association with character information indicating characters included in the character image. Specifically, the attribute value corresponds to information indicating the size of the character included in the character image in the character image, information indicating the relationship with other characters in the vicinity, and the character included in the character image. It includes information indicating the likelihood of the result of the character recognition process.

  In the following description, the verification apparatus 200 has a learning data table 301 and the learning data table 301 stores information in advance as an example. However, the present invention is not limited to this. For example, the verification apparatus 200 does not have the learning data table 301, and the user may input an attribute value to the verification apparatus 200 every time the control unit 400 executes the process.

  Here, an example of information stored in the learning data table 301 will be further described with reference to FIG. FIG. 3 is a diagram illustrating an example of information stored in the learning data table according to the second embodiment. In the example shown in FIG. 3, the case where “recognition result” indicating the character obtained as a result of the character recognition processing is also stored in the character image is shown as an example. In addition, “correct answer” in FIG. 3 indicates characters actually included in the character image. In the following, for the sake of convenience of explanation, the correspondence between “correct answer”, “recognition result”, and “attribute value” is referred to as “example data”.

  Further, in the example shown in FIG. 3, “coordinate”, “distance value”, “correct reading probability”, “morpheme”, “bigram probability”, and “row coordinate” are stored as “attribute values”. It was shown to. Here, “coordinates” and “line coordinates” are information indicating the size of characters included in the character image in the character image. Further, “morpheme” and “bigram probability” are information indicating relevance with other characters in the vicinity (adjacent to the character). The “distance value” and “correct reading probability” are information indicating the likelihood of the result of the character recognition process for characters included in the character image.

  Here, the “distance value” and the “correct reading probability” are obtained, for example, by executing a character recognition process on a character image. The “coordinates” and “line coordinates” can be obtained by identifying the positions of characters and lines in the character image, for example. Further, “morpheme” and “bigram probability” are obtained by, for example, calculating morpheme analysis and bigram probability for a character or character string obtained as a result of character recognition processing.

  Here, the “coordinate” is information indicating the position of the character in the character image. In the example shown in FIG. 3, “x-axis” and “y-axis” are set on the image represented by the character image, and “xs” is a coordinate indicating the upper left point of the character data as “coordinate”. In this example, “ys” and “xe” and “ye” which are coordinates indicating the lower right point of the character data are used.

  The “distance value” is the feature amount calculated for the character included in the character image and the feature amount for the character that is the result of the character recognition processing among the feature amounts in the dictionary table (not shown in FIG. 2). It is information which shows the distance. The “distance value” indicates that the smaller the distance between two feature quantities in the feature space, the higher the reliability of the character recognition result than when the distance is large. In the dictionary table, a feature amount for each character is registered.

  The “correct reading probability” is a value indicating the probability of the character recognition result. The “correct reading probability” indicates that the higher the value, the higher the probability that the character recognition result is correct compared to the case where the value is small. Further, “correct reading probability” indicates that the smaller the value, the lower the probability that the character recognition result is correct compared to the case where the value is high.

  The “morpheme” indicates information on the part of speech obtained for each morpheme as a result of the morpheme analysis. Note that morphological analysis indicates that sentences are divided into meaningful words and parts of speech are discriminated using a dictionary. For example, when morphological analysis is performed on the sentence “I ran”, the analysis results of “I = noun”, “ha = particle”, “run = verb”, “ta = particle”, “. = Punctuation” are obtained. It is done. In the example shown in FIG. 3, “morpheme” is “unregistered word” indicating a morpheme whose part of speech could not be determined using a dictionary, or “registered word” indicating a morpheme whose part of speech could be determined using a dictionary. Or a “symbol”. In the example shown in FIG. 3, morpheme “1” indicates “symbol”, morpheme “2” indicates “registered word”, and morpheme “3” indicates “unregistered word”. Show.

  The case where the part of speech cannot be determined using a dictionary corresponds to the case where words delimited by morphological analysis are not registered in the dictionary. For example, when the noun “Finland” is not registered in the dictionary, the morpheme of each character forming “Finland” is “unregistered word”.

  The “bigram probability” is a linguistic probability, and specifically is a value related to the probability that the character “X2” appears after the character “X1”. For example, taking “Finland” as an example, the bigram probability of “i” indicates the probability of occurrence of “fu” before “i”. For example, in the case where the word “president” often appears, the bigram probability indicating that “large” appears before “torn” is a character other than “torn”, “take”. The value is higher than the bigram probability, which indicates that “large” appears before. In the example illustrated in FIG. 3, the bigram probability is a value calculated by “log (p) * C” using “frequency ratio p” and “constant C”. In the example illustrated in FIG. 3, the “bigram probability” indicates that the closer to “0”, the higher the probability. The “frequency ratio p” indicates the probability that the character “X2” appears after the character “X1”.

  The “line coordinate” is information indicating the position of the line to which the character actually included in the character image belongs. In the example illustrated in FIG. 3, as “line coordinates”, “xs0” and “xe0”, which are coordinates indicating the upper left point of the line to which the character actually included in the character image belongs, and the lower right point of the line. The case where “ys0” and “ye0” which are the coordinates shown are used is shown.

  Here, a specific example of information stored in the learning data table 301 will be further described with reference to FIG. In the example shown in Fig. 3, the case where the character image includes "cardiobike (exercise)" is shown as an example, that is, as shown in the "correct answer" in Fig. 3, the learning data table 301 includes " For each character or symbol included in the cardiobike (exercise), the information obtained by performing the character recognition process is stored.In more detail, as shown in FIG. The recognition result “K” is stored for “K”, and xs “711” ys “2061” xe “747” ye “2101” is stored as “coordinates.” That is, the learning data table 301 is stored in a character image. The character “f” is included, and the character “f” obtained as a result of the character recognition process is stored, and the learning data table 301 has the coordinates of the upper left point of the character “f” included in the character image. xs "711" It is ys “2061”, and the coordinates of the lower right point of “K” are stored as xe “747” ye “2101”.

  For example, the learning data table 301 stores a distance value “666”, a correct reading probability “984”, a morpheme “2”, and a bigram probability “−548112” for the correct answer “K”. In addition, for example, the learning data table 301 stores xs0 “706” ys0 “68” xe0 “774” ye0 “2266” as “line coordinates” for the correct answer “f”. That is, in the learning data table 301, the coordinates indicating the upper left point of the line to which the character “K” included in the character image belongs are xs0 “706” ys0 “68”, and the coordinates indicating the lower right point of the line are xe0 “774” ye0 “2266” is stored. Similarly, the learning data table 301 stores attribute values for “correct answer” after “f”.

  As shown in FIG. 4, the best integrated logical expression table 302 stores the best integrated logical expression for each “combination of similar characters”. FIG. 4 is a diagram illustrating an example of information stored in the best integrated logical expression table according to the second embodiment. The “similar character combination” indicates a combination of characters that tend to be erroneously recognized in the character recognition process, and includes “target character” and “character that is easily misrecognized” as shown in FIG. Here, “target character” indicates a character to be subjected to character recognition processing, and “character that is easily misrecognized” is obtained as a result of erroneous recognition in the character recognition processing for “target character”. Indicates a possible character. Here, “characters that are easily misrecognized” correspond to characters that are similar in shape to the target character. Note that the target character is also referred to as a “first character”, and a character that is easily misrecognized is also referred to as a “second character”.

  For example, a case where the character “I” actually included in the character image is easily misrecognized as “I” will be described as an example. In this case, the “similar character combination” includes the target character “I” and the easily misrecognized character “I”. In addition, as the “similar character combination”, there is a combination of “l” and “1”.

  In the example illustrated in FIG. 4, the best integrated logical expression table 302 associates the character “I” with the best integrated logical expression “(F5 (U = 0.9) orF6 (U = 0.64)) and (F5 (U = 0.76)”. orF6 (U = 0.64) orF2 (U = 2)) and (F5 (U = 0.9) orF7 (U = 0.23)) and (F5 (U = 0.76) orF7 (U = 0.23) orF2 (U = 2)) '' Remember. Note that the details of the best integration logical expression will be described later, and thus the description thereof is omitted here. The best integrated logical formula is also referred to as a “verification formula”.

  As will be described later, the best integrated logical expression table 302 stores combinations of similar characters and best integrated logical expression information by the best integrated logical expression creating unit 404 of the control unit 400. In addition, the best integrated logical expression stored in the best integrated logical expression table 302 is used by the verification unit 406 of the control unit 400.

  The control unit 400 is connected to the input unit 201, the display unit 202, and the storage unit 300. In addition, the control unit 400 has an internal memory that stores a program that defines various control procedures and the like, and executes various control processes. The control unit 400 is, for example, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a central processing unit (CPU), or a micro processing unit (MPU).

  In addition, in the example illustrated in FIG. 2, the control unit 400 includes a reception control unit 401, a learning data acquisition unit 402, a best logical expression creation unit 403, a best integrated logical expression creation unit 404, a character, It has a recognition unit 405 and a verification unit 406.

  The reception control unit 401 receives a combination of similar characters and conditions from the user via the input unit 201. Here, the “condition” is to distinguish a character that is likely to be erroneously recognized as a result of erroneous recognition in the character recognition process for the target character from the target character. For example, the reception control unit 401 displays a display screen as shown in FIG. And the reception control part 401 receives the combination and conditions of a similar character from a user via a display screen as shown in FIG. FIG. 5 is a diagram illustrating an example of the display screen in the second embodiment.

  Here, in FIG. 5, a field 501 indicates a field that receives an input of a target character, and a field 502 indicates a field that receives an input of a character that is easily misrecognized. Further, in the examples shown in the field 501 and the field 502, the case where “I” is input as a target character and “I” is input as a character that is easily misrecognized is shown as an example. A field 503 is a field for receiving a definition used when inputting a condition. In the example shown in the field 503, for example, the case where the definition indicating the correct reading probability is “c” is shown as an example. A field 504 in FIG. 5 indicates a field for accepting an input of a condition. In the example shown in the field 504, the conditions are F1 to F7.

  Here, the conditions F1 to F7 shown in FIG. 5 will be briefly described. Conditions F1 to F7 are examples of conditions input by the user when the target character is “I” and the character that is easily misrecognized is “I”. Below, although demonstrated using the conditions F1-condition F7, this invention is not limited to this, A user may set conditions other than the conditions F1-condition F7 arbitrarily.

  The condition F1 is “c> U”. If the recognition result is correct, the condition F1 indicates that the correct reading probability obtained from the character image to be verified is a predetermined value or more. The condition F1 is a condition based on the fact that the higher the possibility that the recognition result is correct, the higher the correct reading probability. Here, the correct reading probability obtained from the character image to be verified is assigned to “c” in the condition F1. “U” is an indefinite value, and is determined by the best logical expression creation unit 403, which will be described later, using the attribute value.

  The condition F2 is “m = U”. The condition F2 indicates that if the recognition result is correct, the morpheme obtained from the character image to be verified becomes a predetermined morpheme. The condition F2 is a condition based on the fact that if the recognition result is correct, the morphological analysis result for the character that is the recognition result is a predetermined one of “registered word”, “unregistered word”, and “symbol”. A value indicating any one of “registered word”, “unregistered word”, and “symbol” is used for “U”. Here, a value indicating the morpheme obtained from the character image to be verified is substituted for “m” in the condition F2.

  The condition F3 is “dist (Y) −dist (X)> U”. If the recognition result is correct, the condition F3 indicates that the distance value for the target character obtained from the character image to be verified is smaller than the distance value for a character that is likely to be erroneously recognized. The condition F3 is a condition based on the fact that if the recognition result is correct, the distance value for the target character is smaller than the distance value for a character that is easily misrecognized.

  The conditions F1 to F3 are conditions that can be used without being limited to the case where the target character is “I” and the character that is easily misrecognized is “I”. Conditions F4 to F7 are effective conditions when there is a difference in character size. For example, as a case where there is a difference in character size, there is a case where the target character is “I” and a character that is easily misrecognized is “I”.

  The condition F4 is “b (X) −b (Y)> U”. Condition F4 indicates that if the recognition result is correct, the bigram probability obtained from the character image to be verified is larger than the bigram probability of “i”. Condition F4 is a condition based on the fact that “I” is more frequently used than “I”, and the bigram probability of “I” is larger than the bigram probability of “I”.

  The condition F5 is “(ye−ys + 1) / (ye0−ys0 + 1)> U”. If the recognition result is correct, Condition F5 uses the interval on the y-axis of the line coordinates obtained from the character image to be verified as the denominator, and sets the interval on the y-axis of the coordinates obtained from the character image to be verified. It shows that the value made into the molecule | numerator is more than predetermined value. Condition F5 is a character in which “I” is larger in height than “I”, and the value obtained by dividing the height of “I” by the height of the line is the height of “I”. This condition is based on the fact that it is larger than the value divided by the above.

  The condition F6 is “(xe−xs + 1) / (xe0−xs0 + 1)> U”. If the recognition result is correct, the condition F6 uses the interval on the x-axis of the line coordinates obtained from the character image to be verified as the denominator, and the interval on the x-axis of the coordinates obtained from the character image to be verified. It shows that the value made into the molecule | numerator is more than predetermined value. Condition F6 is that “i” has a larger width than “i”, and the value obtained by dividing the width of “i” by the width of the line is the value obtained by dividing the width of “i” by the width of the line. It is a condition based on the fact that it becomes larger than the above.

  The condition F7 is “(ys−ys0 + 1) / (ye0−ys0 + 1)> U”. Condition F7 is that if the recognition result is correct, the interval on the y-axis of the line coordinate obtained from the character image to be verified is used as the denominator, and the y-axis from the upper right point of the coordinate to the upper right point of the line coordinate. It indicates that the value with the distance as a numerator is a predetermined value or more. Condition F7 is that “i” is a character whose height is larger than “i”, and the distance from the upper side of the line to “i” is larger than the distance from the upper side of the line to “i”. It is a condition based on this.

The learning data acquisition unit 402 acquires case data related to a combination of similar characters from the learning data table 301. For example, a case where the target character is “X” and the easily misrecognized character is “Y” will be described as an example. In this case, the learning data acquisition unit 402 acquires case data corresponding to the following (A) to (D) among the case data stored in the learning data table 301.
(A) Case data with correct answer “X” and recognition result “X” (B) Case data with correct answer “X” and recognition result “Y” (C) Correct answer “Y” and recognition result Case data with “Y” (D) Case data with correct answer “Y” and recognition result “X”

  The case where the target character is “I” and the easily misrecognized character is “I” will be further described as an example. The learning data acquisition unit 402 includes case data (A) in which the correct answer is “I” and the recognition result is “I”, and case data (B) in which the correct answer is “I” and the recognition result is “I”. And get. Further, the learning data acquisition unit 402 includes case data (C) in which the correct answer is “I” and the recognition result is “I”, and case data (C) in which the correct answer is “I” and the recognition result is “I”. D). That is, the learning data acquisition unit 402 acquires each case data whose correct answer is “I” or “I” and whose recognition result is “I” or “I”.

  Hereinafter, for convenience of explanation, a set of case data (A) is referred to as “set A”, and a set of case data (B) is referred to as “set B”. The union of “set A” and “set B” is referred to as “set SX”. A set of case data (C) is referred to as “set C”, and a set of case data (D) is referred to as “set D”. The union of “set C” and “set D” is referred to as “set SY”.

  The best logical expression creation unit 403 uses the condition and case data input by the user to distinguish between the target character and the easily misrecognized character for each case data in which the target character is correct. Create a logical expression. Here, the overall image of the process performed by the best logical expression creation unit 403 will be described with reference to FIGS. FIGS. 6A to 6D are diagrams for explaining the overall process performed by the best logical expression creation unit 403 according to the second embodiment.

  In FIGS. 6A to 6D, the attribute values included in each case data are plotted. Specifically, “C” in FIGS. 6-1 to 6-4 indicates case data in which the target character is correct. For example, “C” in FIGS. 6-1 to 6-4 indicates case data whose correct answer is “X”, and each case data included in the “set SX” corresponds. In addition, “E” in FIGS. 6-1 to 6-4 indicates case data in which a character that is easily misrecognized is correct. For example, “E” in FIGS. 6-1 to 6-4 indicates case data whose correct answer is “Y”, and each case data included in the “set SY” corresponds.

  Here, FIG. 6A illustrates an example of case data included in the “set SX” and the “set SY”. Here, as indicated by “arrow” in FIG. 6A, the best logical expression creation unit 403 selects one case data included in the “set SX”. That is, the best logical expression creation unit 403 selects one “C” in FIG. Then, as shown in the group 10 of FIG. 6B, the best logical expression creation unit 403 creates a logical expression that satisfies the selected case data. Then, as shown in groups 11 to 13 in FIG. 6-3, the best logical expression creation unit 403 performs generalization processing on the created logical expressions, thereby indicating the group 10 in FIG. 6-2. Create multiple logical expressions with relaxed conditions than logical expressions. Then, as shown in the group 14 of FIG. 6-4, the best logical expression creation unit 403 selects one logical expression having the highest evaluation value from among the plurality of logical expressions created by executing the generalization process. To do. Further, the best logical expression creation unit 403 executes the processing described with reference to FIGS. 6-1 to 6-4 for each case data included in the “set SX”. A logical expression created by executing the generalization process is also referred to as an “exclusion conditional expression”. The selected logical expression is also referred to as a “high evaluation exclusion condition expression”. Details of the evaluation value will be described later.

  In the following, among the processes by the best logical expression creating unit 403, the process of creating the logical expression shown in FIG. 6-2, the process of creating a plurality of logical expressions by the generalization process shown in FIG. 6-3, The process of selecting one logical expression having the highest evaluation value shown in FIG. 6-4 will be further described.

The process of creating the logical expression shown in FIG. When describing the process of creating a logical expression, the conditions F1 to F7 will be described as examples. Also, for example, the case data includes coordinates “(633, 178), (666, 223)”, distance value “538”, correct reading probability “778”, morpheme “2”, bigram probability “-1489414”, row coordinates The case of “(614, 64), (688, 2278)” is used as an example. Further, for example, it is assumed that the distance value of “i” is “789” and the bigram probability is “−2509827”. For the distance value and bigram probability of “i”, the distance value and bigram probability included in the case data in which “i” is correct are used. Moreover, about the conditions F5-F7, the case where it uses after converting for vertical writing is used for an example. That is, the vertical writing conditions F5 to F7 are as follows.
Condition F5 for vertical writing “(ye−ys + 1) / (xe0−xs0 + 1)> U”
Vertical writing condition F6 “(xe−xs + 1) / (xe0−xs0 + 1)> U”
Condition F7 for vertical writing “(xs−xs0 + 1) / (xe0−xs0 + 1) <U”

  Here, the best logical expression creating unit 403 creates an equation or an inequality related to the attribute value by using the attribute value and the condition included in the case data. For example, based on the fact that the morpheme of the case data is “2” and the morpheme is “2” in many cases of “I”, the best logical expression creation unit 403 sets “m = U” for the condition F2. Set “U” to “2”. As a result, the best logical expression creation unit 403 creates an equation “m = 2” for the condition F2. Similarly, the best logical expression creation unit 403 creates equality and inequalities by setting “U” for other conditions as well.

  Also, the best logical expression creation unit 403 creates a logical expression for the selected case data by combining the created equality and inequality under the AND condition. For example, the equation “m = 2” created for the condition F2 and the equations and inequalities created for other conditions are combined with the and condition. As a result, the best logical expression creation unit 403 creates the following (logical expression 1) as a logical expression.

  (Formula 1) “(c> 778) and (m = 2) and (dist (Y) -dist (X)> 789-538) and (b (X) -b (Y)>-1489414 + 2509827) and ( (Ye-ys + 1) / (xe0-xs0 + 1)> 46/75) and ((xe-xs + 1) / (xe0-xs0 + 1)> 34/75) and ((xs-xs0 + 1) / (xe0-xs0 + 1) <20 / 75) "

(Logical expression 1) may be described as (logical expression 2) below.
(Formula 2) “F1 (U = 778) and F2 (U = 2) and F3 (U = 789-538) and F4 (U = -1489414 + 2509827) and F5 (U = 46/75)) and F6 (U = 34/75) and F7 (U = 20/75) "

  Next, processing for creating a plurality of logical expressions by the generalization processing shown in FIG. 6-3 will be described. The generalization process is a process of relaxing a condition by gradually removing components that form a logical expression and creating a more general logical expression. Specifically, the best logical expression creating unit 403 creates a more general logical expression by combining and reducing the components forming the logical expression. Note that a component means an equation or an inequality included in a logical expression.

  Here, based on the fact that if the number of components is reduced without setting any rules, the number of logical expressions to be created causes a combined explosion, the best logical expression creating unit 403 will have the top three evaluation values below. A method for reducing the number of components in a stepwise manner by limiting to a logical expression will be described as an example. The method described below is referred to as “beam search”. In the following, a case where “beam search” is used will be described as an example. However, the present invention is not limited to this, and other known methods may be used. In the following description, the case of limiting to the top three logical expressions will be described as an example. However, the present invention is not limited to this, and the user may set an arbitrary value.

  Now, the generalization process will be described by taking as an example a case where one component is reduced from the logical expression created in FIG. FIG. 7A is a schematic diagram illustrating generalization processing according to the second embodiment. Here, “601” in FIG. 7A represents the logical expression created in FIG. Further, “1” to “7” in FIG. 7A indicate the equality and inequality created by the best logical expression creation unit 403, respectively.

  Here, in the example indicated by “602” in FIG. 7A, the best logical expression creating unit 403 reduces any one of the components forming the logical expression indicated by “601” in FIG. Thus, “7” logical expressions are created. Note that “x” in “602” in FIG. 7A indicates components reduced by the generalization processing. Then, as indicated by “603” in FIG. 7-1, the best logical expression creating unit 403 includes logical expressions of “3” having the highest evaluation value among the logical expressions of “602” in FIG. Select. Note that the process of selecting a logical expression with a high evaluation value is the same as the process described with reference to FIG.

  Next, with reference to FIG. 7B, an example will be described in which one component is further reduced after one component is reduced from the logical expression created in FIG. 6-2. FIG. 7B is a schematic diagram illustrating generalization processing according to the second embodiment. Here, “604” in FIG. 7-2 represents the logical expression selected in “603” in FIG. 7-2. In addition, “1” to “7” in FIG. 7B indicate equality and inequality created by the best logical expression creation unit 403, respectively.

  Here, as indicated by “604” in FIG. 7B, the best logical expression creating unit 403 sets “3” logical expressions indicated by “603” in FIG. To do. Specifically, the best logical expression creation unit 403 reduces any one of the components of the logical expression for each of “3” logical expressions indicated by “604” in FIG. As a result, in the example indicated by “605” in FIG. 7B, the best logical expression creating unit 403 creates “18” logical expressions. Then, as indicated by “606” in FIG. 7B, the best logical expression creating unit 403 includes the logical expression of “3” having the highest evaluation value among the logical expressions “605” in FIG. Select. Note that the process of selecting a logical expression with a high evaluation value is the same as the process described with reference to FIG.

  Further, the best logical expression creation unit 403 repeats the same processing, thereby reducing the number of components in a stepwise manner from the logical expression created in FIG. 6-2 and creating a plurality of logical expressions. An example of a detailed flow of the generalization process will be described later with reference to FIG.

  Next, processing for selecting one logical expression having the highest evaluation value shown in FIG. 6-4 will be described with reference to FIG. FIG. 8 is a diagram illustrating processing for selecting a logical expression having the highest evaluation value in the second embodiment. “701” in FIG. 8 indicates a logical expression created by the best logical expression creation unit 403. For example, in the example shown in “701” in FIG. 8, “S0” in FIG. 8 represents a logical expression in which no component is reduced, and “S1” in FIG. 8 is reduced by one component. Shows the logical expression.

  Here, as indicated by “702” in FIG. 8, the best logical expression creating unit 403 selects the logical expression having the highest evaluation value among the logical expressions indicated by “701” in FIG. Specifically, the best logical expression creation unit 403 includes the number of case data that should be satisfied by the logical expression (the number of positive cases), the number of case data that should not be satisfied (the number of negative cases), and the components that form the logical expression The evaluation value is determined using the number. Here, the “number of positive cases” indicates how many case data included in the “set SX” have been described. For example, in the example illustrated in FIG. 6-4, the “number of correct cases” is “3”. The “number of negative cases” indicates how many case data included in the “set SY” are satisfied. For example, in the example illustrated in FIG. 6-4, the “number of negative cases” is “0”. The “number of components” indicates the number of “equals and inequalities regarding the attribute of the recognition result” connected using the and condition and the or condition. For example, in the logical expression indicated by “702” in FIG.

  Here, the best logical expression creation unit 403 determines that the smaller the number of negative cases, the higher the evaluation value than the logical expression having a large number of negative cases. In addition, the best logical expression creation unit 403 determines that the larger the number of positive cases, the higher the evaluation value than the logical expression with a smaller number of positive cases. In addition, the best logical expression creation unit 403 determines that the larger the number of components, the higher the evaluation value than the logical expression with a smaller number of components. Then, the best logical expression creation unit 403 selects a logical expression that is determined to have the highest evaluation value as a logical expression. In the following, the logical expression that is determined by the best logical expression creating unit 403 to have the highest evaluation value is referred to as “best logical expression”.

  As described above, the best logical expression creation unit 403 executes the processing described with reference to FIGS. 6-1 to 6-4 for each case data included in the “set SX”. As a result, the best logical expression creation unit 403 creates the best logical expression for each case data included in the “set SX”.

  Returning to the description of FIG. The best integrated logical expression creation unit 404 uses each of the best logical expressions to create a best integrated logical expression that is a logical expression that explains all case data in which the target character is correct. Here, the overall image of the process performed by the best integrated logical expression creation unit 404 will be described with reference to FIGS. FIG. 9A to FIG. 9D are diagrams for explaining the overall image of processing by the best integrated logical expression creation unit in the second embodiment.

  In FIG. 9A to FIG. 9D, the attribute values included in each case data are plotted. Specifically, “C” in FIGS. 9-1 to 9-4 indicates case data in which the target character is correct. For example, “C” in FIGS. 9-1 to 9-4 indicates case data whose correct answer is “X”, and each case data included in the “set SX” corresponds. Further, “E” in FIGS. 9-1 to 9-4 indicates case data in which a character that is easily misrecognized is correct. For example, “E” in FIGS. 9-1 to 9-4 indicates case data whose correct answer is “Y”, and each case data included in the “set SY” corresponds.

  Here, the best integrated logical expression creation unit 404 selects the best logical expression having the highest evaluation value. For example, the best integrated logical expression represented by the group 21 in FIG. Then, the best integrated logical expression creation unit 404 performs a covering check. Specifically, the best integrated logical expression creating unit 404 determines whether there is case data that is not explained (not covered) by the selected best logical expression. Here, the case data targeted in the covering check is case data in which the target character is correct. When it is determined that there is a best integrated logical expression creation unit 404, as shown by “arrow” in FIG. 9-2, one case data that is not covered is selected and shown in the group 22 in FIG. Then, the best logical expression for the selected case data is selected. The best integrated logical expression creation unit 404 combines the best logical expressions selected in FIGS. 9-1 and 9-2 with the or condition. 9-3, the best integrated logical expression creation unit 404 creates a plurality of logical expressions by executing generalization processing on the combined logical expressions. 9-4, the best logical expression creation unit 403 selects one logical expression having the highest evaluation value from the plurality of logical expressions created by executing the generalization process. To do. The best logical expression having the highest evaluation value selected by the best integrated logical expression creating unit 404 is also referred to as “high evaluation conditional expression”.

  Also, the best integrated logical expression creation unit 404 determines whether there is case data that is not explained by the selected logical expression, and if it is determined that there is, the process described with reference to FIGS. 9-2 to 9-4 repeat. On the other hand, when it is determined that there is no case data that is not explained by the selected logical expression, the best integrated logical expression creating unit 404 sets the selected logical expression as the “best integrated logical expression”. As a result, the best integrated logical expression creation unit 404 creates “1” best integrated logical expressions that explain all the case data in which the target character is correct.

  Here, the generalization process by the best integrated logical expression creating unit 404 is the same as the process performed by the best logical expression creating unit 403, and detailed description thereof is omitted. The process of selecting one logical expression having the highest evaluation value by the best integrated logical expression creating unit 404 is the same as the process performed by the best logical expression creating unit 403, and detailed description thereof is omitted.

  As described above, the best logical expression creation unit 403 and the best integrated logical expression creation unit 404 cooperate to perform character recognition using conditions and attribute values for the first character and the second character. If the result of the processing is the first character, a verification formula for verifying the correctness of the result is created.

  When the character recognition unit 405 receives a character image from the input unit 201, the character recognition unit 405 performs character recognition processing on the received character image, and recognizes a character included in the character image as a character code. Then, the character recognition unit 405 calculates a feature amount from the character image of the portion of the character image that includes the character, reads a character having the highest similarity with the calculated feature amount from the dictionary table, and performs character recognition. As a result.

  The verification unit 406 identifies whether the target character stored in the best integrated logical expression table 302 is included in the character recognition result by the character recognition unit 405. For example, the verification unit 406 identifies whether “i” is included in the character recognition result. If the verification unit 406 identifies that it is included, the verification unit 406 reads the best integration logical expression from the best integration logical expression table 302 using the target character included in the character recognition result as a search key, and reads the best integration logical read. Verification is performed using a logical expression.

  The verification process by the verification unit 406 will be described in more detail. The verification unit 406 determines whether the best integrated logical expression read from the best integrated logical expression table 302 satisfies the attribute value related to the target character obtained as the character recognition result. Then, the verification unit 406 determines that the possibility of being correct is high when it is determined that it is satisfied, and determines that the possibility of being incorrect is high when it is determined that it is not satisfied.

  Here, the verification unit 406 uses, for example, the attribute value calculated by the character recognition unit 405 during the character recognition process as the attribute value related to the target character, or calculates the attribute value related to the target character by itself. To use. Hereinafter, an example of a method for the verification unit 406 to calculate the attribute value will be briefly described. Note that the method by which the verification unit 406 calculates the attribute value is not limited to the method described below, and other known methods or methods arbitrarily set by the user may be used.

  For example, when calculating “coordinates” and “line coordinates”, the verification unit 406 identifies the coordinates of characters and lines in the character image. For example, the verification unit 406 identifies the coordinates indicating the upper left point of the character and the coordinates indicating the lower right point of the character, and the coordinates indicating the upper left point of the row and the coordinates indicating the lower right point of the row. Identify.

  For example, when calculating the “distance value”, the verification unit 406 calculates a feature amount for the character included in the character image. In addition, the verification unit 406 acquires a feature amount for a character that is a result of the character recognition process among the feature amounts in the dictionary table. Then, the verification unit 406 calculates a distance value by calculating a distance between the calculated feature value and the acquired feature value.

  Further, for example, when calculating “correct reading probability”, the verification unit 406 calculates a feature amount for a character included in a character image. Also, the verification unit 406 acquires a feature quantity that is the first closest to the calculated feature quantity and a feature quantity that is the second closest to the calculated feature quantity among the feature quantities in the dictionary table. Here, the feature quantity closest to the calculated feature quantity is the feature quantity for the character that is the result of the character recognition process. Then, the verification unit 406 calculates a distance value “d1” between the calculated feature value and the feature value closest to the calculated feature value. Further, the verification unit 406 calculates a distance value “d2” between the calculated feature value and the feature value second closest to the calculated feature value. Then, the verification unit 406 calculates a higher value as the distance value “d1” is smaller than the distance value “d2”, and smaller as the distance value “d1” is larger than the distance value “d2”. The correct reading probability is calculated by calculating the value. That is, as the distance value “d1” is smaller than the distance value “d2”, the feature amount of the character included in the character image is farther from characters other than the character that is the result of the character recognition process. This indicates that the probability of correct reading increases.

  Further, for example, when calculating the “bigram probability”, the verification unit 406 calculates “log (p) * C” using “frequency ratio p” and “constant C”. . Further, for example, when calculating “morpheme”, the verification unit 406 performs morpheme analysis on the character string obtained as a result of the character recognition process, thereby dividing the sentence into meaningful words, and It is calculated by discriminating the part of speech using it.

  The verification unit 406 displays the character recognition result from the display unit 202. Further, the verification unit 406 displays characters from the display unit 202 in a manner different from other characters for characters for which a verification result indicating that there is a high possibility of being incorrect is obtained. For example, the verification unit 406 displays a character that is determined to have a high possibility of being wrong among characters obtained as a result of character recognition using a color different from that of other recognized characters.

  The verification apparatus 200 is realized by using an information processing apparatus such as a known personal computer, workstation, server, mobile phone, PHS (Personal Handyphone System) terminal, mobile communication terminal, or PDA (Personal Digital Assistant). May be. For example, each function of the learning data table 301 and the best integrated logical expression table 302 shown in FIG. 2 is installed in an information processing apparatus such as a PDA. In addition, an information processing apparatus such as a PDA includes a reception control unit 401, a learning data acquisition unit 402, a best logical expression creation unit 403, a best integrated logical expression creation unit 404, a character recognition unit 405, and a verification unit 406. It may be realized by installing each function.

  For example, when the verification apparatus 200 as a server receives a combination of conditions and similar characters from a client, the verification apparatus 200 may create a best integrated logical expression and return it to the client. When the verification apparatus 200 as a server receives a character recognition result from the client, the verification apparatus 200 may perform verification using a verification formula and then return the verification result to the client.

[Processing by Verification Device According to Second Embodiment]
Next, processing by the verification apparatus 200 according to the second embodiment will be described. In the following, a case where the target character is “I” and a character that is easily misrecognized as “I” will be described as an example unless otherwise specified.

[Best integration formula creation process]
First, an example of the flow of the best integrated logical expression creation process in the second embodiment will be described with reference to FIG. FIG. 10 is a flowchart illustrating an example of the flow of best integrated logical expression creation processing according to the second embodiment.

  As illustrated in FIG. 10, when the reception control unit 401 receives a combination of similar characters and a condition (Yes in step S101), the learning data acquisition unit 402 displays case data related to the combination of similar characters as learning data. Obtained from the table 301 (step S102). For example, the learning data acquisition unit 402 acquires each case data whose correct answer is “I” or “I” and whose recognition result is “I” or “I”.

  The best logical expression creation unit 403 creates the best logical expression for each case data in which the target character is correct (step S103). Note that a detailed example of the process flow for creating the best logical expression will be described later with reference to FIG.

  Then, the best integrated logical expression creation unit 404 selects the best logical expression having the highest evaluation value from the best logical expressions created for each case data in which the target character is correct (step S104). Then, the best integrated logical expression creation unit 404 performs a cover check (step S105). Specifically, the best integrated logical expression creating unit 404 determines whether there is case data that is not explained (not covered) by the selected best logical expression.

  Here, when the best integrated logical expression creation unit 404 determines that there is case data that is not explained (not covered) by the selected best logical expression (Yes in step S106), the best logic for the case data that has not been covered is determined. The expressions are integrated under the or condition (step S107). That is, the best integrated logical expression creation unit 404 selects the case data that has not been covered, and obtains the best logical expression for the selected case data and the best logical expression that has been subjected to the covering check in step S105 described above. Combine with conditions.

  Then, the best integrated logical expression creation unit 404 executes a generalization process on the logical expression created by combining with the or condition (step S108). As a result, the best integrated logical expression creation unit 404 creates a plurality of logical expressions by executing generalization processing. A detailed example of the flow of the generalization process will be described with reference to FIG.

  Then, the best integrated logical expression creation unit 404 selects the logical expression having the highest evaluation value (step S109). That is, the best integrated logical expression creation unit 404 selects a logical expression having the highest evaluation value from among a plurality of logical expressions created by executing the generalization process. An example of a detailed flow of processing for selecting a logical expression having the highest evaluation value will be described later with reference to FIG.

  Then, the best integrated logical expression creation unit 404 performs a covering check on the selected logical expression (step S105). The best integrated logical expression creation unit 404 repeats the above steps S105 to S108 until it determines that there is no case data that is not explained (not covered) by the selected best logical expression (No at step S106).

  On the other hand, a case will be described where, in step S106 described above, the best integrated logical expression creation unit 404 determines that there is no case data that is not explained (not covered) by the selected best logical expression (No in step S106). In this case, the best integrated logical expression creation unit 404 selects the logical expression that is the target of the coverage check as the “best integrated logical expression” and stores it in the best integrated logical expression table 302 in association with the target character ( Step S110).

[Best formula creation process]
Next, an example of the flow of the best logical expression creation process in the second embodiment will be described with reference to FIG. FIG. 11 is a flowchart illustrating an example of the flow of the best logical expression creation process in the second embodiment. Note that the processing flow described with reference to FIG. 11 corresponds to step S103 in FIG.

  As shown in FIG. 11, the best logical expression creation unit 403 selects one case data in which the target character is correct (step S201). For example, the best logical expression creation unit 403 selects one case data in which “I” is correct. Then, the best logical expression creating unit 403 creates a logical expression that satisfies the selected case data (step S202). For example, the best logical expression creation unit 403 creates an equality or inequality for each condition input by the user using the selected attribute value, and creates the new equality or inequality for each condition under the and condition. Create logical expressions by combining them.

  Then, the best logical expression creation unit 403 performs generalization processing on the logical expressions created by combining with the AND condition (step S203). As a result, the best logical expression creation unit 403 creates a plurality of logical expressions by executing generalization processing. A detailed example of the flow of the generalization process will be described with reference to FIG.

  Then, the best logical expression creating unit 403 selects a logical expression having the highest evaluation value (step S204). That is, the best logical expression creation unit 403 selects a logical expression having the highest evaluation value from among a plurality of logical expressions created by executing the generalization process. Here, the logical expression selected by the best logical expression creation unit 403 becomes the “best logical expression”. An example of a detailed flow of processing for selecting a logical expression having the highest evaluation value will be described later with reference to FIG.

  Thereafter, the best logical expression creation unit 403 determines whether there is unprocessed case data (step S205). That is, the best logical expression creation unit 403 determines whether the best logical expression has been created for all case data in which the target character is correct. If the best logical expression creation unit 403 determines that there is unprocessed case data (Yes at step S205), the process returns to step S201 described above and repeats the process. On the other hand, when determining that there is no unprocessed case data (No at Step S205), the best logical expression creating unit 403 ends the best logical expression creating process and starts Step S104 in the process of FIG.

  As a result, when the best logical expression creating process by the best logical expression creating unit 403 is completed, the best logical expression is created for each case data in which the target character is correct.

[Generalization processing]
Next, an example of the flow of generalization processing in the second embodiment will be described with reference to FIG. FIG. 12 is a flowchart for explaining an example of the flow of generalization processing in the second embodiment. Note that the processing flow described with reference to FIG. 12 corresponds to step S108 in FIG. 10 and step S203 in FIG.

  In the following, a case where the best logical expression creation unit 403 executes the generalization process in step S203 of FIG. 11 will be described as an example. In the following, for convenience of explanation, the number of components forming the logical expression to be generalized, that is, the number of components of the logical expression created in step S202 of FIG. 11 is “n”. A case will be described as an example.

  As shown in FIG. 12, the best logical expression creation unit 403 sets “1” for the parameter “i” (step S301), and reduces the number of components forming the logical expression to be generalized by one. Each logical expression is created (step S302). Here, the number of components of each logical expression created in step S302 is “n−1”. For example, when the number of components is “7”, the best logical expression creation unit 403 reduces the number of components by one by reducing any one of the seven equalities or inequalities to “6”. “7” logical expressions are created.

  Then, the best logical expression creation unit 403 selects, from each of the created logical expressions, logical expressions that fall in the upper “k” having the highest evaluation value (step S303). For example, the best logical expression creation unit 403 selects the top “3” logical expressions having the highest evaluation value from the “7” logical expressions created in step S302.

  Then, the best logical expression creation unit 403 adds “1” to the parameter “i” (step S304). For example, the case where the parameter “i” is “1” will be described as an example. The best logical expression creation unit 403 adds “1” to the parameter “1” to obtain “2”.

  The best logical expression creation unit 403 compares the parameter “i” with the number of components “n” and determines whether the parameter “i” is smaller than the number of components “n” (step S305). For example, when the parameter is “2” and the number of components is “7”, the best logical expression creation unit 403 determines that the number is small. If the parameter is “7” and the number of components is “7”, the best logical expression creation unit 403 determines that the parameter is not small.

  Here, when the best logical expression creation unit 403 determines that the parameter “i” is smaller than the number of components “n” (Yes in Step S305), For each selected logical expression, each logical expression in which the number of components forming the logical expression is reduced by one is created (step S306). For example, a case will be described as an example where “3” logical expressions are selected in step S303 described above, and the number of components of “3” logical expressions is “6”. In this case, the best logical expression creation unit 403 reduces the number of constituent elements by one by reducing any one of six equalities or inequalities for each of “3” logical expressions, thereby reducing “5”. Create “6” logical expressions. That is, the best logical expression creation unit 403 creates “6 × 3 = 18” logical expressions.

  Then, the best logical expression creation unit 403 selects a logical expression that falls within the top “k” having the highest evaluation value for the created logical expression (step S303). That is, for example, the best logical expression creating unit 403 selects the top “3” logical expressions having the highest evaluation value from among the “18” logical expressions created. The best logical expression creation unit 403 repeats the process until it is determined in step S305 described above that the parameter “i” is not smaller than the number of components “n” (steps S303 to S306).

  On the other hand, if it is determined in step S305 described above that the parameter “i” is not smaller than the number of components “n” (No in step S305), the generalization process ends.

[Process to select the logical expression with the highest evaluation value]
Next, an example of a processing flow for selecting a logical expression having the highest best evaluation value in the second embodiment will be described with reference to FIG. FIG. 13 is a flowchart illustrating an example of a processing flow for selecting a logical expression having the highest evaluation value in the second embodiment. Note that the processing flow described with reference to FIG. 13 corresponds to step S109 in FIG. 10 and step S204 in FIG. In the following, a case where the best logical expression creation unit 403 selects the best logical expression in step S204 of FIG. 11 will be described as an example.

  As shown in FIG. 13, the best logical expression creation unit 403 identifies the number of positive cases, the number of negative cases, and the number of components for each logical expression (step S401). Then, the best logical expression creation unit 403 determines whether there is a logical expression having a negative case of “0” (step S402). Here, when the best logical expression creating unit 403 determines that there is a logical expression having a negative case of “0” (Yes in step S402), the logical expression having a negative case of “0” is calculated as (number of positive cases / (Number of components) is calculated for each logical expression (step S403). Then, the best logical expression creation unit 403 selects the logical expression having the highest (number of positive cases / number of components) as the best logical expression (step S404).

  On the other hand, when it is determined that there is no logical expression whose negative example is “0” (No in step S402), the best logical expression creating unit 403 calculates (positive case number × component number) / (negative case number) as a logical expression. It calculates for every formula (step S405). The best logical expression creation unit 403 selects the logical expression having the highest (number of positive cases × number of components) / (number of negative cases) as the best logical expression (step S406).

[Verification processing]
Next, an example of the flow of verification processing according to the second embodiment will be described with reference to FIG. FIG. 14 is a flowchart illustrating an example of the flow of verification processing according to the second embodiment.

  As illustrated in FIG. 14, when the character recognition unit 405 receives a character image (Yes in step S501), the character recognition unit 405 executes character recognition processing (step S502). Specifically, the character recognition unit 405 calculates a feature amount from a portion of the character image that includes the character, reads a character having the highest similarity with the calculated feature amount from the dictionary table, and performs character recognition. As a result.

  Then, the verification unit 406 determines whether the character recognition result by the character recognition unit 405 includes the target character stored in the best integrated logical expression table 302 (step S503). If the verification unit 406 determines that the target character is not included (No at step S503), the verification unit 406 displays the character recognition result on the display unit 202 as it is (step S504).

  On the other hand, the case where it is determined that the target character is included in the character recognition result by the character recognition unit 405 will be described (Yes in step S503). In this case, the verification unit 406 reads the best integrated logical expression from the best integrated logical expression table 302 using the target character included in the character recognition result as a search key (step S505). For example, when “I” is included in the character recognition result, the best integrated logical expression associated with the target character “I” is read. Then, the verification unit 406 executes verification processing using the read best integrated logical expression (step S506).

  Then, the verification unit 406 displays, from the display unit 202, a character for which a verification result indicating that there is a high possibility of being wrong is obtained in a manner different from other characters (step S507). For example, the verification unit 406 displays a character that is determined to have a high possibility of being wrong among characters obtained as a result of character recognition using a color different from that of other recognized characters.

[Effect of Example 2]
As described above, according to the second embodiment, the verification apparatus 200 creates a verification formula using conditions and attribute values. In addition, when a character image is input, the verification apparatus 200 performs character recognition processing on the input character image. Then, the verification apparatus 200 identifies whether the target character is included in the result of the character recognition process, and performs verification using the verification formula when it is identified as included. As a result, according to the second embodiment, even if the character obtained as the character recognition result is a similar character having other similar characters, it is accurately determined whether the character obtained as the character recognition result is correct. It can be verified.

  Further, according to the second embodiment, the verification apparatus 200 receives an input of a plurality of conditions for distinguishing the second character and the first character for each attribute value of the target character, and receives the logical expression Create Then, the verification apparatus 200 creates a plurality of logical expressions by generalizing the created logical expressions for each attribute value for the target character. In addition, the verification apparatus 200 increases the value for each of the plurality of logical expressions as the number of attribute values satisfied by the logical expression among the attribute values for the target character increases. An evaluation value that is higher as the number of attribute values that the logical expression satisfies among the attribute values is smaller is calculated. Then, the verification apparatus 200 selects a logical expression having the highest calculated evaluation value. Here, the logical expression selected by the verification device 200 is the best logical expression. Further, the verification apparatus 200 selects one logical expression having the highest evaluation value among the logical expressions selected for each attribute value. Then, the verification apparatus 200 determines whether or not the selected logical expression satisfies all the attribute values for the target character. If it is determined that the selected logical expression is satisfied, the selected logical expression is set as the best integrated logical expression. decide. If the verification apparatus 200 determines that the attribute is not satisfied, the verification apparatus 200 integrates the selected logical expression with respect to the other attribute values that are not satisfied, and determines the best integrated logical expression. As a result, it is possible to accurately verify whether the character obtained as the character recognition result is correct.

  In addition, according to the second embodiment, the verification apparatus 200 determines that other characters obtained as a result of the character recognition process when the attribute value related to the character included in the character image does not satisfy the verification formula as a result of the verification. As a result of output from the display unit 202 in a different manner, the user can easily grasp similar characters that are likely to be erroneously recognized.

  Although the embodiments of the present invention have been described so far, the present invention may be implemented in other embodiments besides the above-described embodiments. Therefore, other embodiments will be described below.

[Learning data acquisition unit]
For example, in the above-described embodiment, when the target character is “I” and the easily misrecognized character is “I”, the learning data acquisition unit 402 determines whether the correct answer is “I” or “I”. ”And the case data whose recognition result is“ I ”or“ I ”is acquired as an example. However, the present invention is not limited to this. For example, the learning data acquisition unit 402 may acquire case data whose correct answer is “I” or “I” regardless of the recognition result.

[Characters that are easily misrecognized]
Further, for example, in the above-described embodiment, the case where there is one character that is easily misrecognized corresponding to one target character has been described as an example. However, the present invention is not limited to this. For example, there may be two or more easily misrecognized characters corresponding to one target character.

[Verification processing]
Further, for example, in the above-described embodiment, the case where the verification is performed using the verification formula created for the target character has been described, but the present invention is not limited to this. For example, if “i” is included in the result of the character recognition process, the verification device verifies using not only the verification formula created for “i” but also the validation formula created for “i”. May be. The verification device may verify that the result of the character recognition process is correct when the verification formula for “I” is satisfied and the verification formula for “I” is not satisfied.

[System configuration]
Also, among the processes described in this embodiment, all or part of the processes described as being performed automatically can be performed manually, or the processes described as being performed manually can be performed. All or a part can be automatically performed by a known method. In addition, the processing procedure, control procedure, specific name, and information including various data and parameters shown in the above-described document and drawings can be arbitrarily changed unless otherwise specified.

  Further, each component of each illustrated apparatus is functionally conceptual, and does not necessarily need to be physically configured as illustrated. That is, the specific form of distribution / integration of each device is not limited to the one shown in the figure, and all or a part of the distribution / integration may be functionally or physically distributed in arbitrary units according to various loads or usage conditions. Can be integrated and configured.

  For example, to explain with reference to the example shown in FIG. 2, the first device having the best integrated logical expression table 302, the character recognition unit 405, and the verification unit 406, the learning data table 301, the reception control unit 401, and the learning The second device having the data acquisition unit 402, the best logical formula creation unit 403, and the best integrated logical formula creation unit 404 may be a separate device.

  In this case, when the second device creates the best integrated logical expression, it transmits it to the first device. Further, when the first device receives the best integrated logical expression from the first device, the first device stores it in the best integrated logical expression table 302. Further, the first device executes a verification process using the best integrated logical expression.

[Computer]
The various processes described in the above embodiments can be realized by executing a prepared program on a computer such as a personal computer or a workstation. In the following, an example of a computer that executes a verification program having the same function as that of the above-described embodiment will be described with reference to FIG. FIG. 15 is a schematic diagram illustrating an example of a computer that executes a verification program according to the second embodiment.

  As illustrated in FIG. 15, the computer 3000 according to the second embodiment includes a keyboard 3001, a microphone 3002, a speaker 3003, and a display 3004. The computer 3000 further includes a communication unit 3006, a CPU 3010, a ROM 3011, an HDD (Hard Disk Drive) 3012, and a RAM (Random Access Memory) 3013. In the computer 3000, each unit is connected by a bus 3009 or the like.

  As shown in FIG. 15, the ROM 3011 stores in advance a reception control program 3011a, a learning data acquisition program 3011b, and a best logical expression creation program 3011c. The ROM 3011 further stores in advance a best integrated logical expression creation program 3011d, a character recognition program 3011e, and a verification program 3011f. Here, the reception control program 3011a is a control program that exhibits the same function as the reception control unit 401 shown in the second embodiment. The learning data acquisition program 3011b is a control program that exhibits the same function as the learning data acquisition unit 402. The best logical expression creation program 3011c is a control program that exhibits the same function as the best logical expression creation unit 403. The best integrated logical expression creation program 3011d is a control program that exhibits the same function as the best integrated logical expression creation unit 404. The character recognition program 3011e is a control program that exhibits the same functions as the character recognition unit 405 and the best integrated logical expression creation unit 404. The verification program 3011f is a control program that exhibits the same functions as the verification unit 406 and the best integrated logical expression creation unit 404. Note that these programs 3011a to 3011f may be integrated or separated as appropriate, similarly to each component of the verification apparatus 200 shown in FIG.

  Then, the CPU 3010 reads out these programs 3011a to 3011f from the ROM 3011 and executes them, and as shown in FIG. 15, for each program 3011a to 3011f, an acceptance control process 3010a, a learning data acquisition process 3010b, It functions as the best logical expression creation process 3010c, the best integrated logical expression creation process 3010d, the character recognition process 3010e, and the verification process 3010f. Each of the processes 3010a to 3010f includes the reception control unit 401, the learning data acquisition unit 402, the best logical expression creation unit 403, the best integrated logical expression creation unit 404, and the character recognition unit 405 shown in FIG. And the verification unit 406 respectively.

  The HDD 3012 is provided with a learning data table 3012a and a best integrated logical expression table 3012b. Each table 3012a to 3012b corresponds to the learning data table 301 and the best integrated logical expression table 302 shown in FIG.

  Then, the CPU 3010 reads the learning data table 3012a and the best integrated logical expression table 3012b and stores them in the RAM 3013. The learning data data 3013a, the best integrated logical expression data 3013b, and the condition data 3013c stored in the RAM 3013 are read out. And the verification program is executed using the best logical expression data 3013d.

[Others]
The verification program described in the present embodiment can be distributed via a network such as the Internet. The verification program can also be executed by being recorded on a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM, an MO, and a DVD, and being read from the recording medium by the computer.

DESCRIPTION OF SYMBOLS 100 Verification apparatus 200 Verification apparatus 201 Input part 202 Display part 300 Storage part 301 Learning data table 302 Best integration logical expression table 400 Control part 401 Reception control part 402 Learning data acquisition part 403 Best logical expression creation part 404 Best integration logical expression Creation unit 405 Character recognition unit 406 Verification unit

Claims (5)

Receiving an input of a plurality of conditions for distinguishing between the first character and the second character that may be obtained as a result of erroneous recognition in the character recognition process for the first character; An attribute value, information indicating the size of the character included in the character image of the character in the character image, information indicating the relationship between the character and other nearby characters, and the character A conditional expression creating unit that creates a conditional expression for each attribute value including a plurality of pieces of information including information indicating the probability of the result of the character recognition process;
For each conditional expression created by the conditional expression creating section, an exclusion conditional expression creating section that creates a plurality of exclusion conditional expressions excluding at least one of the conditions included in the conditional expression;
For each exclusion condition expression created for each conditional expression by the exclusion condition expression creation unit, the higher the number of attribute values that the exclusion condition expression satisfies among the attribute values for the first character, the higher the value The evaluation value that becomes higher as the number of attribute values that the exclusion conditional expression satisfies among the attribute values for the second character is smaller is calculated, and the calculated evaluation value is the highest. A selection unit that selects a high evaluation exclusion conditional expression that is a conditional expression for each conditional expression;
From the high evaluation exclusion conditional expressions selected for each conditional expression by the selection unit, one high evaluation conditional expression having the highest evaluation value is selected, and whether all the attribute values for the first character are satisfied If it is determined and satisfied that the result of character recognition processing of the character included in the character image is the first character, the high evaluation condition selected as the verification expression for verifying the correctness of the result When determining a formula as the verification formula and determining that the formula is not satisfied, a determination unit that determines the verification formula after being integrated with a high evaluation condition formula for other attribute values that are not satisfied,
Identifying whether the first character of the result of character recognition processing contained, if identified as containing, a verification unit for performing verification using the verification equation created by the determination unit A verification apparatus characterized by comprising. If it is determined that incorrect results of verification by the verification unit, claims and further comprising a output unit for outputting the display unit in a manner different from the other characters in the character recognition process 1 The verification device described in 1. Computer
Receiving an input of a plurality of conditions for distinguishing between the first character and the second character that may be obtained as a result of erroneous recognition in the character recognition process for the first character; An attribute value, information indicating the size of the character included in the character image of the character in the character image, information indicating the relationship between the character and other nearby characters, and the character A conditional expression creating step for creating a conditional expression for each attribute value including a plurality of pieces of information and information indicating the probability of the result of the character recognition process;
For each conditional expression created by the conditional expression creating step, an exclusion conditional expression creating step for creating a plurality of exclusion conditional expressions excluding at least one of the conditions included in the conditional expression;
For each exclusion conditional expression created for each conditional expression in the exclusion conditional expression creation step, the higher the number of attribute values that the exclusion conditional expression satisfies among the attribute values for the first character, the higher the value The evaluation value that becomes higher as the number of attribute values that the exclusion conditional expression satisfies among the attribute values for the second character is smaller is calculated, and the calculated evaluation value is the highest. A selection process for selecting a high evaluation exclusion conditional expression that is a conditional expression for each conditional expression,
From the high evaluation exclusion conditional expressions selected for each conditional expression in the selection step, one high evaluation conditional expression having the highest evaluation value is selected, and whether all attribute values for the first character are satisfied If it is determined and satisfied that the result of character recognition processing of the character included in the character image is the first character, the high evaluation condition selected as the verification expression for verifying the correctness of the result When determining a formula as the verification formula and determining that the formula is not satisfied, a determination step of determining the verification formula after integrating with a high evaluation condition formula for other attribute values that are not satisfied,
Identifying whether the first character of the result of character recognition processing contained, if identified as containing, a verification step of performing verification using the verification equation created by the determination step A verification method characterized by performing. Receiving an input of a plurality of conditions for distinguishing between the first character and the second character that may be obtained as a result of erroneous recognition in the character recognition process for the first character; An attribute value, information indicating the size of the character included in the character image of the character in the character image, information indicating the relationship between the character and other nearby characters, and the character A conditional expression creation procedure for creating a conditional expression for each attribute value including a plurality of pieces of information including information indicating the probability of the result of the character recognition process;
For each conditional expression created by the conditional expression creating procedure, an exclusion conditional expression creating procedure for creating a plurality of exclusion conditional expressions excluding at least one of the conditions included in the conditional expression;
For each exclusion conditional expression created for each conditional expression by the exclusion conditional expression creation procedure, the higher the number of attribute values that the exclusion conditional expression satisfies among the attribute values for the first character, the higher the value The evaluation value that becomes higher as the number of attribute values that the exclusion conditional expression satisfies among the attribute values for the second character is smaller is calculated, and the calculated evaluation value is the highest. A selection procedure for selecting a high evaluation exclusion conditional expression that is a conditional expression for each conditional expression,
Of the high evaluation exclusion conditional expressions selected for each conditional expression by the selection procedure, one high evaluation conditional expression with the highest evaluation value is selected, and whether all the attribute values for the first character are satisfied If it is determined and satisfied that the result of character recognition processing of the character included in the character image is the first character, the high evaluation condition selected as the verification expression for verifying the correctness of the result When determining a formula as the verification formula and determining that the formula is not satisfied, a determination procedure for determining the verification formula after being integrated with a high evaluation condition formula for other attribute values that are not satisfied,
A verification procedure for identifying whether or not the first character is included in the result of the character recognition process, and performing verification using the verification formula created by the pre- decision procedure when the first character is identified. A verification program characterized in that it is executed. Information indicating the size of the character contained in the character image in the character image, information indicating the relationship between the character and other nearby characters, and the probability of the result of character recognition processing for the character An attribute value storage unit that stores an attribute value including at least one of the information to be displayed in association with character information indicating a character included in the character image;
A receiving unit that receives a condition for distinguishing the second character from the first character that may be obtained as a result of erroneous recognition in the character recognition process for the first character;
A conditional expression creating unit that creates a conditional expression for each attribute value stored in the attribute value storage unit in association with character information indicating the first character;
For each conditional expression created by the conditional expression creating section, an exclusion conditional expression creating section that creates a plurality of exclusion conditional expressions excluding at least one of the conditions included in the conditional expression;
For each exclusion condition expression created for each conditional expression by the exclusion condition expression creation unit, the higher the number of attribute values that the exclusion condition expression satisfies among the attribute values for the first character, the higher the value And the higher the smaller the number of attribute values that the exclusion condition expression satisfies among the attribute values stored in the attribute value storage unit in association with the character information indicating the second character, the higher the value. A selection unit that calculates an evaluation value, and selects a high evaluation exclusion conditional expression that is an exclusion conditional expression with the highest calculated evaluation value for each conditional expression;
From the high evaluation exclusion conditional expressions selected for each conditional expression by the selection unit, one high evaluation conditional expression having the highest evaluation value is selected, and whether all the attribute values for the first character are satisfied If it is determined and satisfied that the result of character recognition processing of the character included in the character image is the first character, the high evaluation condition selected as the verification expression for verifying the correctness of the result When the formula is determined as the verification formula and it is determined that the formula is not satisfied, the determination section includes a determination unit that determines the verification formula after being integrated with a high evaluation condition formula for other attribute values that are not satisfied A creation device characterized by that.

Images