JP2011081454A - Information processing apparatus, information processing method, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information processing apparatus, along with an information processing method, program and recording medium, capable of correcting an error in a recognized character string to a correct character, without using dictionaries other than recognition dictionaries or word dictionaries. <P>SOLUTION: A matching part 18, based on the similarity calculated from the feature value of a character image and the feature value of characters in a character recognition dictionary 14, extracts candidate characters from the first to the fifth candidates. A character string generation unit 19 generates a recognized result character string consisting of the first candidate characters. A word string generation unit 21 generates a word character string using a word dictionary 24 from the recognized result character string. A replacement decision unit 22 compares the recognized result character string and the word character string and determines whether the character should be replaced or not based on the coincidence of the characters and the similarity of the characters. A replacing part 23 replaces the character determined to be replaced by the replacement decision unit 22 to the character in the word character string and transmits and outputs the replaced recognized result character string to an output device 3. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an information processing apparatus, an information processing method, a program, and a recording medium that optically read a document on which characters such as sentences are described, perform character recognition, and output the recognized characters. The present invention relates to an information processing apparatus, an information processing method, a program, and a recording medium that can be corrected.

  In the field of optical character recognition, an information processing device such as an optical character reader that optically reads a document on which characters such as sentences are written, performs character recognition, and outputs the recognized character uses language processing. It employs a word correct technology that corrects character recognition errors and improves recognition accuracy by performing post-processing.

  As a word correction technique, for example, a word dictionary containing many words is prepared, and a character string composed of recognized candidate characters (hereinafter referred to as “recognized words”) is stored in the word dictionary. A technique is generally used in which it is determined whether or not the words match, and when the words match, the word is regarded as correct and the recognized word is replaced with the word matched in the word dictionary.

  However, simply replacing a recognized word with a word recorded in the word dictionary means that if the recognition result is correct but the word is not recorded in the word dictionary, the recognized word matches the word recorded in the word dictionary. There is a possibility that it will be replaced with an incorrect correction, that is, an incorrect word.

  In order to prevent the recognition rate from being reduced due to erroneous correction, a word is created from the recognized candidate characters, a word similar to the created word or a completely matching word is searched from the word dictionary, and the similarity between the searched word and the recognized word A method has been proposed in which the recognition word is output or replaced with a word in the word dictionary by calculating (see Patent Document 1, for example).

  For each recognized character, a plurality of candidate characters are extracted, such as the first to fifth candidate characters, and a plurality of candidate characters are combined when a word is formed from the recognized characters. A method of replacing a correct word with higher accuracy by forming a word has been proposed (for example, see Patent Document 2). However, even in this method, there is a possibility that a character that can form a correct word is not included in the plurality of candidate characters, and there is a case where correct replacement cannot be performed.

  As a method of replacing a correct word even when there is no correct character among a plurality of candidate characters, a method of replacing it with a correct word using a similar character dictionary in which similar characters are classified has been proposed. In this method, when there is a character that matches a character in the similarity character dictionary among a plurality of candidate characters, a similar character belonging to the classification is added to the candidate character. Then, a word is formed from candidate characters to which similar characters are added, the formed word is compared with a word in the word dictionary, and the matching word is output as a correct character string (see, for example, Patent Document 3).

JP 10-134150 A JP 58-4490 A Japanese Patent Laid-Open No. 5-135211

  However, when using a similar character dictionary like the method described in Patent Document 3, it is necessary to store the similar character dictionary in a storage device such as a memory in addition to the character recognition dictionary for recognizing characters. The storage capacity of the storage device increases. Also, depending on how many similar characters are recorded for each character recorded in the dictionary, too few characters make sense, so more characters must be recorded. There is a problem that a large storage capacity is required to store the dictionary.

  Furthermore, the characters extracted from the similarity character dictionary do not necessarily include the correct characters, and all of them are added as candidate characters, and are similar or completely coincide with the words existing in the word dictionary. Searching for and replacing a word complicates the processing and increases the processing time.

  An object of the present invention is to provide an information processing apparatus, an information processing method, a program, and a recording medium that can correct an error of a recognized character to a correct character without using a dictionary other than a character recognition dictionary and a word dictionary. It is.

The present invention relates to a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character;
A character string row cutout unit that cuts out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout unit that cuts out one character image representing each character from the character string image cut out by the character string line cutout unit;
A feature extraction unit that extracts a feature amount of a character represented by each character image from the character image cut out by the character cutout unit;
The feature amount extracted by the feature extraction unit is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection unit for selecting at least one from the character recognition dictionary;
A character string generation unit that generates a recognition result character string by combining candidate characters for each character image selected by the selection unit;
The candidate characters constituting the recognition result character string generated by the character string generation unit are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation unit that extracts a character string or a word that matches only some characters and generates a word character string composed of the extracted words;
If there is a different character by comparing the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at the corresponding position, feature extraction is performed for the different characters The feature amount extracted by the unit is compared with the feature amount stored in association with the character that matches the character of the word character string among the character feature amounts recorded in the character recognition dictionary stored in the storage unit. A replacement determination unit that determines to replace a character in the recognition result character string with a corresponding character in the word character string when the comparison result satisfies a predetermined comparison criterion;
An information processing apparatus comprising: a replacement unit that replaces a character in a recognition result character string determined to be replaced by a replacement determination unit with a character at a corresponding position in a word character string.

The present invention also includes a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character,
A character string row cutout unit that cuts out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout unit that cuts out one character image representing each character from the character string image cut out by the character string line cutout unit;
A feature extraction unit that extracts a feature amount of a character represented by each character image from the character image cut out by the character cutout unit;
The feature amount extracted by the feature extraction unit is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection unit for selecting at least one from the character recognition dictionary;
A character string generation unit that generates a recognition result character string by combining candidate characters for each character image selected by the selection unit;
The candidate characters constituting the recognition result character string generated by the character string generation unit are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation unit that extracts a character string or a word that matches only some characters and generates a word character string composed of the extracted words;
When there is a different character by comparing the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at the corresponding position, the storage unit Among the feature quantities recorded in the character recognition dictionary stored in the memory, and the feature quantities stored in association with the characters matching the characters in the recognition result character string, and the features recorded in the character recognition dictionary stored in the storage unit The feature amount stored in association with the character that matches the character of the word character string is compared, and when the comparison result satisfies a predetermined comparison criterion, the character in the recognition result character string is A replacement determination unit that determines to replace with a corresponding character;
An information processing apparatus comprising: a replacement unit that replaces a character in a recognition result character string determined to be replaced by a replacement determination unit with a corresponding character in a word character string.

  In the present invention, the replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position. If there are different characters, for the different characters, the character that matches the character of the word character string out of the feature amount extracted by the feature extraction unit and the feature amount recorded in the character recognition dictionary stored in the storage unit The similarity is calculated by comparing with the feature quantity stored in association with each other, and when the calculated similarity is equal to or greater than a predetermined threshold, the character in the recognition result character string is replaced with the corresponding character in the word character string. It is characterized by determining to replace.

  In the present invention, the replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position. If there are different characters, for the different characters, the feature amount stored in association with the character that matches the character of the recognition result character string among the feature amounts recorded in the character recognition dictionary stored in the storage unit, and The similarity is calculated by comparing the feature amount recorded in the character recognition dictionary stored in the storage unit with the feature amount stored in association with the character that matches the character of the word character string, and the calculated similarity When the degree is equal to or greater than a predetermined threshold, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string.

  Further, the present invention provides the replacement determination unit, when the calculated similarity is less than a predetermined threshold, and when the word string generation unit generates the word character string, When a word string that matches all characters, or a word that matches only some characters and is not included in the word character string, the calculated similarity is less than a predetermined threshold value. When the value is equal to or greater than a predetermined second threshold value, the character string in the recognition result character string is determined to be replaced with the corresponding character in the word character string.

  In the present invention, when the calculated similarity is less than a predetermined threshold, and the number of characters of the word including the different characters is equal to or greater than a predetermined reference number, the calculated similarity is When it is equal to or more than a second predetermined threshold value that is a value less than a predetermined threshold value, it is determined that a character in the recognition result character string is replaced with a corresponding character in the word character string.

Further, the present invention provides an information processing including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character. An information processing method for recognizing characters using a character recognition dictionary and a word dictionary stored in a storage unit in an apparatus,
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
If there are different characters by comparing the recognition result character string generated in the character string generation process and the word character string generated in the word string generation process for each character at the corresponding position, feature extraction is performed for the different characters The feature amount extracted in the process is compared with the feature amount stored in association with the character matching the character of the word character string among the character feature amounts recorded in the character recognition dictionary stored in the storage unit. A replacement determination step for determining that a character in the recognition result character string is replaced with a corresponding character in the word character string when the comparison result satisfies a predetermined comparison criterion;
A replacement step of replacing a character in the recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in the word character string.

Further, the present invention provides an information processing including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character. An information processing method for recognizing characters using a character recognition dictionary and a word dictionary stored in a storage unit in an apparatus,
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
When there are different characters by comparing the recognition result character string generated in the character string generation step and the word character string generated in the word string generation step for each character at the corresponding position, the storage unit Among the feature quantities recorded in the character recognition dictionary stored in the memory, and the feature quantities stored in association with the characters matching the characters in the recognition result character string, and the characters recorded in the character recognition dictionary stored in the storage unit When the comparison result meets a predetermined comparison criterion, the character in the recognition result character string is converted to the word character string. A replacement determination step for determining to replace with the corresponding character in
A replacement step of replacing a character in the recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in the word character string.

In addition, the present invention provides a computer including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character. ,
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
If there are different characters by comparing the recognition result character string generated in the character string generation process and the word character string generated in the word string generation process for each character at the corresponding position, feature extraction is performed for the different characters The feature amount extracted in the process is compared with the feature amount stored in association with the character matching the character of the word character string among the character feature amounts recorded in the character recognition dictionary stored in the storage unit. A replacement determination step for determining that a character in the recognition result character string is replaced with a corresponding character in the word character string when the comparison result satisfies a predetermined comparison criterion;
This is a program for executing a replacement step of replacing a character in a recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in a word character string.

In addition, the present invention provides a computer including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character. ,
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
When there are different characters by comparing the recognition result character string generated in the character string generation step and the word character string generated in the word string generation step for each character at the corresponding position, the storage unit Among the feature quantities recorded in the character recognition dictionary stored in the memory, and the feature quantities stored in association with the characters matching the characters in the recognition result character string, and the characters recorded in the character recognition dictionary stored in the storage unit When the comparison result meets a predetermined comparison criterion, the character in the recognition result character string is converted to the word character string. A replacement determination step for determining to replace with the corresponding character in
This is a program for executing a replacement step of replacing a character in a recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in a word character string.
The present invention is also a computer-readable recording medium on which the program is recorded.

  According to the present invention, the storage unit stores a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a word dictionary that records a word composed of at least one character. The character string line cutout unit cuts out a character string image representing a character string for each line from an image of a document in which the character string line that is a character string is described. The character cutout unit cuts out one character image representing each character from the character string image cut out by the character string row cutout unit. The feature extraction unit extracts a feature amount of a character represented by each character image from the character image cut out by the character cutout unit. The selection unit compares the feature amount extracted by the feature extraction unit with the feature amount recorded in the character recognition dictionary stored in the storage unit, and each character represented by the character image is recognized as a character represented by the character image. At least one candidate character to be obtained is selected from the character recognition dictionary. The character string generation unit generates a recognition result character string by combining candidate characters for each character image selected by the selection unit. The word string generation unit collates the candidate characters constituting the recognition result character string generated by the character string generation unit with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary And a character string that matches all characters, or a word that matches only some characters, and generates a word character string composed of the extracted words. The replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position, and if there is a different character, a different character For the feature amount extracted by the feature extraction unit and the feature amount stored in association with the character that matches the character of the word character string among the character feature amounts recorded in the character recognition dictionary stored in the storage unit When the comparison result satisfies a predetermined comparison criterion, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string. Then, the replacement unit replaces the character in the recognition result character string determined to be replaced by the replacement determination unit with the character at the corresponding position in the word character string.

  Therefore, it is possible to correct the recognized character error to a correct character without using a dictionary other than the character recognition dictionary and the word dictionary. Since no dictionary other than the character recognition dictionary and the word dictionary is required, an increase in storage capacity can be avoided.

  Further, according to the present invention, the storage unit stores a character recognition dictionary that records a character and a feature value obtained by quantifying the character feature, and a word dictionary that records a word composed of at least one character. . The character string line cutout unit cuts out a character string image representing a character string for each line from an image of a document in which the character string line that is a character string is described. The character cutout unit cuts out one character image representing each character from the character string image cut out by the character string row cutout unit. The feature extraction unit extracts a feature amount of a character represented by each character image from the character image cut out by the character cutout unit. The selection unit compares the feature amount extracted by the feature extraction unit with the feature amount recorded in the character recognition dictionary stored in the storage unit, and each character represented by the character image is recognized as a character represented by the character image. At least one candidate character to be obtained is selected from the character recognition dictionary. The character string generation unit generates a recognition result character string by combining candidate characters for each character image selected by the selection unit. The word string generation unit collates the candidate characters constituting the recognition result character string generated by the character string generation unit with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary And a character string that matches all characters, or a word that matches only some characters, and generates a word character string composed of the extracted words. The replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position, and if there is a different character, a different character For the feature quantity recorded in the character recognition dictionary stored in the storage unit, the feature quantity stored in association with the character that matches the character of the recognition result character string, and the character recognition dictionary stored in the storage unit Is compared with the feature quantity stored in association with the character that matches the character of the word character string, and when the comparison result satisfies a predetermined comparison criterion, It is determined that the character is replaced with the corresponding character in the word string. Then, the replacement unit replaces the character in the recognition result character string determined to be replaced by the replacement determination unit with the character at the corresponding position in the word character string.

  Therefore, it is possible to correct the recognized character error to a correct character without using a dictionary other than the character recognition dictionary and the word dictionary. Since no dictionary other than the character recognition dictionary and the word dictionary is required, an increase in storage capacity can be avoided.

  According to the invention, the replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position. If there are different characters, the different character matches the character of the word character string among the feature amount extracted by the feature extraction unit and the feature amount recorded in the character recognition dictionary stored in the storage unit. The similarity is calculated by comparing with the feature value stored in association with the character to be matched, and when the calculated similarity is equal to or greater than a predetermined threshold, the character in the recognition result character string corresponds to the word character string. Judged to replace with a character. Therefore, since the similarity based on the feature amount is used as an index, the similarity of characters can be determined with a simple calculation.

  According to the invention, the replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position. If there is a different character, the feature value stored for the different character is associated with the character that matches the character of the recognition result character string among the feature values recorded in the character recognition dictionary stored in the storage unit. And the feature quantity recorded in the character recognition dictionary stored in the storage unit in association with the character that matches the character of the word character string to calculate the degree of similarity. When the similarity is equal to or greater than a predetermined threshold, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string. Therefore, since the similarity based on the feature amount is used as an index, the similarity of characters can be determined with a simple calculation.

  According to the invention, the replacement determination unit is recorded in the word dictionary when the calculated similarity is less than a predetermined threshold, and when the word string generation unit generates the word character string. When a word string that matches all characters or a word that matches only some characters and is not included in the word character string, the calculated similarity is determined in advance. When the value is equal to or greater than a predetermined second threshold value that is less than the threshold value, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string.

  Therefore, when only one word candidate is generated, the word candidate is likely to be correct and can be replaced as an optimum word even if the similarity is less than a predetermined threshold. . For example, in optical character recognition, if one character of a word consisting of a plurality of characters is not correctly recognized due to “faint” or “smear”, the degree of similarity will be low and will be less than a predetermined threshold. However, if there is only one word extracted from the word dictionary, there is a high possibility that it is correct, and there is an effect that it can be corrected to a correct character by correcting using this word. .

  According to the present invention, the replacement determination unit determines that the calculated similarity is greater than the predetermined reference number of characters including the different characters when the calculated similarity is less than a predetermined threshold. When the degree is equal to or greater than a predetermined second threshold value that is a value less than a predetermined threshold value, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string. Therefore, for words with a large number of characters, even if the degree of similarity is less than a predetermined threshold value, there is a high possibility that the word is correct, and an optimal word can be selected and replaced.

  In addition, according to the present invention, a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character are included. In the information processing apparatus, when a character is recognized using the character recognition dictionary and the word dictionary stored in the storage unit, a character string is extracted from an image of a document in which a character string row that is a character string is described in the character string row cutout process. A character string image representing is cut out for each line. In the character cutout step, a character image representing each character is cut out character by character from the character string image cut out in the character string row cutout step. In the feature extraction step, the feature amount of the character represented by each character image is extracted from the character image cut out in the character cutout step. In the selection step, the feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and each character represented by the character image is recognized as a character represented by the character image. At least one candidate character to be obtained is selected from the character recognition dictionary. In the character string generation step, a recognition result character string is generated by combining candidate characters for each character image selected in the selection step. In the word string generation step, the candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary And a character string that matches all characters, or a word that matches only some characters, and generates a word character string composed of the extracted words. In the replacement determination step, if there is a different character by comparing the recognition result character string generated in the character string generation step and the word character string generated in the word string generation step for each character at the corresponding position, a different character About the feature amount extracted in the feature extraction step and the feature stored in association with the character that matches the character of the word character string out of the feature amount of the character recorded in the character recognition dictionary stored in the storage unit When the comparison result satisfies a predetermined comparison criterion, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string. In the replacement step, the character in the recognition result character string determined to be replaced in the replacement determination step is replaced with the character at the corresponding position in the word character string.

  Therefore, it is possible to correct the recognized character error to a correct character without using a dictionary other than the character recognition dictionary and the word dictionary. Since no dictionary other than the character recognition dictionary and word dictionary is required, an increase in the storage capacity of the information processing apparatus can be avoided.

  In addition, according to the present invention, a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character are included. In the information processing apparatus, when a character is recognized using the character recognition dictionary and the word dictionary stored in the storage unit, a character string is extracted from an image of a document in which a character string row that is a character string is described in the character string row cutout process. A character string image representing is cut out for each line. In the character cutout step, a character image representing each character is cut out character by character from the character string image cut out in the character string row cutout step. In the feature extraction step, the feature amount of the character represented by each character image is extracted from the character image cut out in the character cutout step. In the selection step, the feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and each character represented by the character image is recognized as a character represented by the character image. At least one candidate character to be obtained is selected from the character recognition dictionary. In the character string generation step, a recognition result character string is generated by combining candidate characters for each character image selected in the selection step. In the word string generation step, the candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary And a character string that matches all characters, or a word that matches only some characters, and generates a word character string composed of the extracted words. In the replacement determination step, if there is a different character by comparing the recognition result character string generated in the character string generation step and the word character string generated in the word string generation step for each character at the corresponding position, a different character For the feature quantity recorded in the character recognition dictionary stored in the storage unit, the feature quantity stored in association with the character that matches the character of the recognition result character string, and the character recognition dictionary stored in the storage unit Is compared with the feature quantity stored in association with the character that matches the character of the word character string, and when the comparison result satisfies a predetermined comparison criterion, It is determined that the character is replaced with the corresponding character in the word string. In the replacement step, the character in the recognition result character string determined to be replaced in the replacement determination step is replaced with the character at the corresponding position in the word character string.

  Therefore, it is possible to correct the recognized character error to a correct character without using a dictionary other than the character recognition dictionary and the word dictionary. Since no dictionary other than the character recognition dictionary and word dictionary is required, an increase in the storage capacity of the information processing apparatus can be avoided.

  Moreover, according to this invention, it can provide as a program for making a computer perform each process of an information processing method.

  Furthermore, according to the present invention, it is possible to provide a computer-readable recording medium that records a program for causing a computer to execute each step of the information processing method.

It is a block diagram which shows the structure of the information processing apparatus 1 which is one Embodiment of this invention. It is a figure which shows an example of the image 40 which the image data received from the input device 2 represents. It is a figure which shows an example of the character string line image 41 cut out by the character string line cutout part. It is a figure which shows an example of the character image 42 cut out by the character cutout part 16. FIG. It is a figure for demonstrating the feature-value extracted by the feature extraction part. It is a figure which shows an example of the candidate character 43 and the recognition result character string 44. FIG. It is a figure which shows an example of the word character string 45 produced | generated by the word sequence production | generation part 21. FIG. It is a figure which shows the comparative example 46 of a recognition result character string and a word character string. It is a figure which shows an example of the feature-value of the character of a word character string, and the feature-value of the candidate character in a corresponding position. 4 is a flowchart illustrating a processing procedure of character recognition processing executed by the information processing device 1, the input device 2, and the output device 3. 4 is a flowchart illustrating a processing procedure of a first post-processing executed by the post-processing unit 20. It is a figure which shows an example of the character image 52 cut out by the character cutout part. It is a figure which shows an example of the candidate character 53 and the recognition result character string 54. FIG. It is a figure which shows an example of the word character string 55 produced | generated by the word sequence production | generation part 21. FIG. It is a figure which shows the comparative example 56 of a recognition result character string and a word character string. It is a figure which shows an example of the feature-value of the character of a word character string, and the feature-value of the candidate character in a corresponding position. It is a flowchart which shows the process sequence of the 2nd post-process which the post-processing part 20a performs. It is a figure which shows an example of the image 60 which the image data received from the input device 2 represents. It is a figure which shows an example of the character string line image 61 cut out by the character string line cutout part. It is a figure which shows an example of the character image 62 cut out by the character cutout part 16. FIG.

It is a figure which shows an example of the candidate character 63 and the recognition result character string 64. FIG. It is a figure which shows an example of the word character string 65 produced | generated by the word sequence production | generation part 21. FIG. It is a figure which shows the comparative example 66 of a recognition result character string and a word character string. It is a figure which shows an example of the feature-value of the character of a word character string, and the feature-value of the candidate character in a corresponding position. It is a flowchart which shows the process sequence of the 3rd post-process which the post-processing part 20b performs.

  FIG. 1 is a block diagram showing the configuration of the information processing apparatus 1 according to the first embodiment of the present invention. The information processing method according to the present invention is executed by the information processing apparatus 1. The information processing device 1 is connected to the input device 2 and the output device 3, reads a document on which text or the like is written as an image by the input device 2, performs character recognition based on image data representing the read image, and recognizes the character This is a device that corrects an error in a character string consisting of characters into a correct character string and outputs the corrected character string to the output device 3.

  The input device 2 is an image reading device such as a scanner that reads a document as an image, and transmits image data representing the read image to the information processing device 1. The output device 3 is a display device such as a monitor configured by a liquid crystal display, for example, and displays information received from the information processing device 1 on a screen. In the embodiment illustrated in FIG. 1, the input device 2 and the output device 3 are configured as devices independent of the information processing device 1, but may be included in the information processing device 1.

  The information processing apparatus 1 is configured by a computer such as a personal computer, for example, and includes a control unit 11, an input / output control unit 12, a memory 13, a character recognition dictionary 14, a character string line cutout unit 15, a character cutout unit 16, and a feature extraction unit 17. , A matching unit 18, a character string generation unit 19, and a post-processing unit 20. The post-processing unit 20 includes a word string generation unit 21, a replacement determination unit 22, a replacement unit 23, and a word dictionary 24.

  The control part 11 is comprised by the central processing unit (Central Processing Unit; abbreviation CPU) which is not shown in figure, for example. The input / output control unit 12 controls transmission / reception of information between the control unit 11 and the input device 2 and transmission / reception of information between the control unit 11 and the output device 3. The memory 13 as a storage unit is configured by a storage device such as a semiconductor memory device or a hard disk device, and stores a program executed by the control unit 11 and information necessary for executing the program. Information stored in the memory 13 is read and written by the control unit 11. The control unit 11 controls the input / output control unit 12 by executing a program stored in the memory 13, as well as a character string line cutout unit 15, a character cutout unit 16, a feature extraction unit 17, a matching unit 18, a character Functions such as the column generation unit 19 and the post-processing unit 20 are realized. In the present embodiment, the memory 13 is included in the information processing apparatus 1, but may be configured by an independent storage device connected to the information processing apparatus 1, or may be configured by a removable recording medium. .

  The character recognition dictionary 14 is a dictionary that records a character represented by a character code and a feature value obtained by quantifying the character feature in association with each other. The word dictionary 24 is a dictionary that records words composed of at least one character. The word dictionary 24 also stores character codes for each character. The character recognition dictionary 14 and the word dictionary 24 are stored in the memory 13.

  The control unit 11 receives the image data transmitted from the input device 2 by the input / output control unit 12 and sends the received image data to the character string line cutout unit 15. The character string row cutout unit 15 cuts out a character string row image representing a character string from the image represented by the image data received from the control unit 11 for each line. The character cutout unit 16 cuts out a character image that is a rectangular image representing characters one by one from the character string row image cut out by the character string row cutout unit 15. The feature extraction unit 17 extracts a feature amount obtained by quantifying the character feature represented by the character image cut out by the character cutout unit 16.

  The matching unit 18 serving as a selection unit performs matching processing with the character recognition dictionary 14. The matching process is a process of comparing the characteristics of the character image with the characteristics of the characters recorded in the character recognition dictionary 14 and quantifying how much they are similar. Specifically, the matching unit 18 compares the feature amount extracted by the feature extraction unit 17 with the feature amount of all the characters recorded in the character recognition dictionary 14 to determine the character and character represented by the character image. The similarity with the characters recorded in the recognition dictionary 14 is calculated. Then, based on the calculated similarity, at least one candidate character of the character represented by the character image, for example, a first to fifth candidate character is extracted. The character string generation unit 19 generates a recognition result character string in which the candidate characters of the first candidate are arranged in order from the left among the candidate characters indicated by the candidate characters extracted by the matching unit 18.

  The post-processing unit 20 performs a first post-processing. In the first post-processing, the word string generation unit 21 combines a plurality of candidate characters of each character constituting the recognition result character string generated by the character string generation unit 19 and collates with the word dictionary 24 to obtain the optimum A word string composed by selecting a word is generated. The replacement determination unit 22 compares the word character string generated by the word string generation unit 21 with the recognition result character string generated by the character string generation unit 19, and when the comparison result satisfies a predetermined comparison criterion, Is determined to be replaced. The replacement unit 23 replaces the character in the recognition result character string with the character in the word character string for the character determined to be replaced by the replacement determination unit 22, and outputs the replaced recognition result character string to the output device 3. Hereinafter, the character string row cutout unit 15, the character cutout unit 16, the feature extraction unit 17, the matching unit 18, the character string generation unit 19, and the post-processing unit 20 will be described in detail.

  FIG. 2 is a diagram illustrating an example of the image 40 represented by the image data received from the input device 2. An image 40 represented by image data received from the input device 2 is an image obtained by reading a document on which a character string row is written by the input device 2. The character string line is a character string composed of at least one line. The character string may be written horizontally or vertically. In horizontal writing, the horizontal direction is a character string, and the vertical direction is a line. In vertical writing, the vertical direction is a character string and the horizontal direction is a line. The image may be a color image, but in the present embodiment, the following description will be made assuming that the image is a monochrome image. The portion composed of white pixels is the background portion, and the portion composed of black pixels is the character portion.

  In the example of the image 40 shown in FIG. 2, a character string line extending over two lines is shown. The first line indicates “A questionnaire about the resident's insurance medical system.” The line says, “We held a briefing session in advance to gain cooperation and understanding.”

  FIG. 3 is a diagram illustrating an example of the character string row image 41 cut out by the character string row cutout unit 15. The character string line segmentation unit 15 receives image data received from the input device 2 by the input / output control unit 12 by the control unit 11 from the control unit 11. A character string row image representing a character string is cut out for each row from the image represented by the received image data.

  The method for cutting out the character string row image may be a well-known method. In the present embodiment, for example, the character string row cutout unit 15 determines from the distribution of blank portions, that is, white pixel portions, and character portions, that is, black pixel portions in the image. Cut out the character string line image. Specifically, a character string row image is cut out by utilizing the fact that there is a certain distribution of white pixels between rows. The method of cutting out the character string line image is not limited to this method, and any other method may be used.

  In the example of the character string row image 41 illustrated in FIG. 3, two character string row images of the character string row images 411 and 412 are illustrated. The character string line image 411 is an image in which the character string “about the resident's insurance medical system” is displayed, and the character string line image 412 is “an explanatory meeting was performed in advance to obtain cooperation and understanding”. Is an image on which a character string “.” Is displayed.

  FIG. 4 is a diagram illustrating an example of the character image 42 cut out by the character cutout unit 16. The character cutout unit 16 cuts out characters one by one from the character string row image cut out by the character string row cutout unit 15 as a character image that is a rectangular image.

  A well-known method may be used as the character cutout method, provided that the character is cut out in a circumscribed rectangle with the coordinates of the most protruding portions in the vertical and horizontal directions as the maximum value. In the present embodiment, the coordinates are coordinates in a coordinate system in which, for example, the upper left vertex of the image 40 illustrated in FIG. 2 is the origin, the left and right directions toward the image 40 are the X axis, and the vertical direction is the Y axis. The character image is a rectangular image surrounded by four sides including two sides parallel to the X axis and two sides parallel to the Y axis, and is a rectangular image having a minimum size including one character.

  In the example of the character image 42 shown in FIG. 4, a character image of 15 characters is shown. The first character image 421 is a character image obtained by cutting out the character “ON”, and “I”, “person”, “NO”, “K”, “K”, “medicine”, “ Character images of the characters “medicine”, “control”, “degree”, “ni”, “tsu”, “i”, “te” and “,” are shown.

  FIG. 5 is a diagram for explaining the feature amount extracted by the feature extraction unit 17. The feature extraction unit 17 extracts a feature amount obtained by quantifying the character feature represented by the character image cut out by the character cutout unit 16. The feature extraction unit 17 stores the extracted feature amount in the memory 13 for each character image for the post-processing unit 20.

  As a feature extraction method, a method called “mesh feature” is well known, and in this embodiment, the feature extraction unit 17 extracts a feature amount using a method called “mesh feature”. “Mesh feature” is a process of dividing a character image, which is a cut-out rectangular image, into meshes, and quantifying the number of black pixels in each mesh for normalization.

  FIG. 5A is a diagram in which the character image 30 of the character “A” is divided into 8 × 8 = 64 meshes 31 and divided. FIG. 5B is a diagram illustrating an example of the feature data 32 obtained by quantifying and normalizing the number of black pixels in each mesh 31 illustrated in FIG. The feature data 32 is data representing the feature amount of the character portion of each mesh 31 for each 8 × 8 mesh 33 at the corresponding position. The corresponding position means that they are at the same position in the 8 × 8 mesh.

  For example, each mesh 33 in the uppermost row indicates that the feature amount of the first mesh 31 is “0” and the feature amount of the second mesh 31 is “0” in order from the left. , The feature amount of the third mesh 31 is “15”, the feature amount of the fourth mesh 31 is “34”, and the feature amount of the fifth mesh 31 is “ 0 ”, the feature amount of the sixth mesh 31 is“ 0 ”, the feature amount of the seventh mesh 31 is“ 3 ”, and the feature amount of the eighth mesh 31 is It indicates that the feature amount is “0”.

  FIG. 6 is a diagram illustrating an example of the candidate character 43 and the recognition result character string 44. FIG. 6A shows an example of the candidate character 43 extracted by the matching unit 18. The candidate character 43 is a candidate character extracted by the matching unit 18 as a character represented by the character image 42. The matching unit 18 compares the feature amount extracted by the feature extraction unit 17 with the feature amount of all characters recorded in the character recognition dictionary 14 and records them in the character recognition dictionary 14 and the character represented by the character image. The degree of similarity with the calculated character is calculated, and at least one candidate character for the character represented by the character image is extracted based on the calculated degree of similarity. In the present embodiment, a maximum of five candidate characters, a first candidate to a fifth candidate character, are extracted from characters with high similarity.

  The similarity is an index for evaluating how similar the character represented by the character image and the character recorded in the character recognition dictionary 14 are, and indicates that the similarity is higher as the similarity is higher. In the present embodiment, the degree of similarity is evaluated using an index called simple similarity as the similarity. The simple similarity S is calculated by equation (1).

Here, the feature data of the character indicated by the character image is X, the feature data of the character recorded in the character recognition dictionary 14 is Y, the feature amount of the mth mesh of the character image is x _m , and the character recognition dictionary 14 The feature amount of the m-th mesh of the recorded characters is y _m , and the number of meshes is k. In the present embodiment, the m-th mesh is an 8 × 8 mesh, so k = 64, m = 1, 2, 3,... 8 in order from the left in the uppermost stage, up to the right end. When coming, m = 9, 10, 11,... From the left end of the next stage, and m = 64 at the right end of the lowermost stage.

  The rightmost denominator of equation (1) becomes constant when the character feature data represented by the character image and the character feature data recorded in the character recognition dictionary 14 are normalized. It is only necessary to perform calculations for molecules. In this embodiment, since the mesh feature is used for an 8 × 8 mesh, the numerator can be expressed as shown in Equation (2).

  If the expression (2) is used, the value of the calculation result becomes very large and difficult to handle. In this embodiment, the value obtained by dividing the calculation result of the expression (2) by “512” as shown in the expression (3) Is used as the simple similarity S.

According to Expression (3), the maximum value of the simple similarity S is set to be “128”.
The feature extraction unit 17 calculates the similarity with the characters recorded in the character recognition dictionary 14 for each character image cut out by the character cutout unit 16 according to Expression (3). Then, from the characters recorded in the character recognition dictionary 14, up to five candidate characters from the first candidate to the fifth candidate are extracted in descending order of similarity. A candidate character is a character that is a candidate for the character represented by the character image. If the similarity is less than a predetermined minimum standard, it is not extracted as a candidate character. The predetermined minimum standard is, for example, “80”. In the present embodiment, the predetermined minimum standard is “80”, but is not limited to this value. Alternatively, the first to fifth candidates may always be extracted without making a determination based on a predetermined minimum standard. In the present embodiment, the candidate characters are up to the fifth candidate. However, the present invention is not limited to this, and for example, up to the tenth candidate may be extracted. However, it should be noted that the amount of data to be stored increases when the character string becomes longer or when the number of recognized character strings increases.

  In the candidate character 43 shown in FIG. 6A, input characters are arranged in order from the left in the top row, the first candidate candidate character for each character in the top row in the second row, and the first character in the third row. Second candidate candidate character for each character in the upper row, Third candidate character for each uppermost character in the fourth row, Fourth candidate candidate character for each uppermost character in the fifth row In the sixth row, the fifth candidate character for each character in the uppermost row is shown. The input character is the original character of the candidate character, that is, the character described in the manuscript, and is hereinafter also referred to as “correct answer character”. The number attached to the lower side of each candidate character represents the similarity of the candidate character. The first candidate is a candidate character having the highest similarity, and thereafter becomes a candidate character of the second candidate, the third candidate, the fourth candidate, and the fifth candidate in descending order of similarity.

  In the example of the candidate character 43 shown in FIG. 6A, “input”, “residence”, “person”, “no”, “protection”, “drug”, “medicine” are input characters at the top. , “Medicine”, “control”, “degree”, “ni”, “tsu”, “i”, “te” and “,” are shown. For the first to fifth candidates, for example, for the correct character “ON”, the first candidate “ON 121”, the second candidate “person 120”, the third candidate “large 115”, and the fourth candidate “λ 110”. And the fifth candidate is blank.

  FIG. 6B shows an example of the recognition result character string 44 generated by the character string generation unit 19. The character string generation unit 19 generates a recognition result character string 44 in which the candidate characters of the first candidate are arranged in order from the left among the candidate characters indicated by the candidate character 43 extracted by the matching unit 18. The character string generation unit 19 stores candidate characters 43 in the memory 13 for the post-processing unit 20. In the example shown in FIG. 6B, the recognition result character string 44 is a character string “Regarding the resident blue tremor medical system”. When the processing by the post-processing unit 20 is not performed, the recognition result character string 44 is output as a recognition result. In the present embodiment, thereafter, the first post-processing is executed by the post-processing unit 20.

  The post-processing unit 20 executes a first post-processing. That is, a word character configured by combining a plurality of candidate characters of each character constituting the recognition result character string 44 generated by the character string generation unit 19 and collating with the word dictionary 24 and selecting an optimum word Generate a column. Then, the generated word character string and the recognition result character string 44 generated by the character string generation unit 19 are compared to correct an error in the recognition result character string 44, and the corrected recognition result character string is output. 3 is output. The first post-processing is executed by the word string generation unit 21, the replacement determination unit 22, and the replacement unit 23 included in the post-processing unit 20.

  Specifically, the word string generation unit 21 includes a character string generated by combining a plurality of candidate characters of each character constituting the recognition result character string 44 generated by the character string generation unit 19, and the word dictionary 24. The word candidate is generated by collating with the words recorded in, and a word character string formed by selecting an optimum word from the generated word candidates is generated.

  A method for generating word candidates from candidate characters includes a method using morphological analysis, and any method may be used. In the present embodiment, in order from the first character of the recognition result character string 44, for each candidate character, a character string that completely matches or partially matches a word recorded in the word dictionary 24 is a word. Generate as a candidate.

  For example, in the recognition result character string 44 shown in FIG. 6B, the character string of the first character to the third character is “occupied blue”. Here, if the word “resident” exists in the word dictionary 24, this “resident” is extracted as a word candidate corresponding to “resident blue”. Further, if the word “resident” is recorded in the word dictionary 24, “resident” is also a word candidate.

  If “resident” and “resident” are extracted from the word dictionary 24 as word candidates, the third candidate “blue” in the recognition result character string 44 corresponds to the first candidate candidate character “blue”. If there is a character “person” among the candidate characters of the second candidate to the fifth candidate, a character constituting the word “resident” may be generated as a candidate character of the character constituting the recognition result character string 44. it can. The candidate characters of the second candidate to the fifth candidate corresponding to the candidate character of the first candidate are the candidate characters of the second candidate to the fifth candidate whose input characters are the same as the candidate characters of the first candidate. Then, since the word candidate “resident” has a longer word length than the word candidate “resident”, the “resident” having the longer word length is selected as the optimum word.

  If the character “person” does not exist among the candidate characters of the second to fifth candidates corresponding to the candidate character “blue”, that is, as shown in FIG. 6A, the recognition result If the number of characters matching the characters constituting the character string 44 is the same, the word having the larger number of characters constituting the word is selected. That is, in both the word candidate “resident” and the word candidate “resident”, the character string “resident” matches the first and second character strings of the recognition result character string 44. . Since the word candidate “resident” has 2 characters, the word “resident” has 3 characters. Therefore, “resident” is selected. . This is an example of selecting a partially matching character string as an optimum word. Hereinafter, partial matching is also referred to as “partial matching”.

  For the fourth character of the recognition result character string 44, the word string generation unit 21 sets a character string that completely or partially matches a word recorded in the word dictionary 24 as an optimum word. select. The word string generation unit 21 repeats this process until the optimum word for the last character of the recognition result character string 44, for example, the character “,” in the example shown in FIG. 6B is selected.

  However, in this process, the word dictionary 24 is collated using a combination of candidate characters other than the candidate characters constituting all the recognition result character strings 44, and partial matching with the words recorded in the word dictionary 24 is allowed. Since it is configured, it is necessary to consider the following.

  For example, the fifth and sixth characters constituting the recognition result character string 44 shown in FIG. 6B are correctly the character string “insurance”, but the matching result shows the fifth and fifth characters of the recognition result character string 44. The sixth character is “Short”, and when the word dictionary 24 is searched, words such as “insurance” and “transfer” that partially match are generated as word candidates. That is, in the candidate character 43 shown in FIG. 6A, the characters “K” and “K” are not present in the “K” candidate character and the “K” candidate character, respectively. There are no candidates, and a plurality of word candidates that are partially matched are generated. For example, a word that partially matches such as “San-in” that is partially matched with “Yin”, which is the third candidate for “Steep”, and “Hill” that is partially matched with “Ling” that is the second candidate for “Steep”. Is recorded in the word dictionary 24, it is generated as a word candidate. An example of the word candidates generated in this way is shown in FIG.

  FIG. 7 is a diagram illustrating an example of the word character string 45 generated by the word string generation unit 21. In the example of the word character string 45 shown in FIG. 7, an optimum word such as a word 451 and a plurality of generated word candidates such as a word candidate 452 are mixed. In the word candidate 452, four words “insurance”, “transfer”, “sanin”, and “hill” are generated. When there are a plurality of candidates, the word string generation unit 21 assigns priorities according to the following three conditions, selects an optimal word, and generates a word character string including only the optimal word. Hereinafter, a case where a word candidate consists of two characters will be described as an example.

  Condition 1 is a condition in which the first character is a character that exists in the candidate character and the second character is a character that exists in the candidate character. Condition 2 is that the first character is a character that exists in the candidate character and the second character is a character that does not exist in the candidate character, or the first character is in the candidate character. The condition is that the character does not exist and the second character is a character that exists in the candidate character. Condition 3 is a condition that the similarity between two characters is the maximum. The similarity between two characters is the similarity between a word candidate and two characters at positions corresponding to the word candidate among the characters of the recognition result character string. The corresponding position means that the recognition result character string and the word character string have the same order from the first character. The similarity between two characters is an average value of the similarity between the characters constituting the two characters. In this embodiment, the average value of the similarity of two characters is used, but the total value of the similarity of two characters may be used.

  If there is one word candidate satisfying the condition 1 among a plurality of word candidates, the word string generation unit 21 selects the word candidate as the optimum word. If there is no word candidate satisfying the condition 1 and there is one word candidate satisfying the condition 2, the word candidate is selected as the optimum word. When there are a plurality of word candidates satisfying the condition 1, or when there are a plurality of word candidates satisfying the condition 2, a word candidate satisfying the condition 3 is selected as an optimum word.

  For example, in the case of the word candidate 452 of the fifth and sixth characters shown in FIG. 7, “insurance”, “transfer”, “San-in”, and “hill” are generated as word candidates. None of these word candidates satisfies condition 1 but satisfies condition 2, and therefore, an optimal word is selected according to condition 3. As for the respective similarities, “insurance” is “118”, “transfer” is “117”, “sanin” is “114”, and “hill” is “116”. Therefore, the word string generation unit 21 selects “insurance” having the maximum similarity “118” from among these four word candidates as the optimum word.

  In the present embodiment, the optimum word is selected according to the above-described conditions. However, the candidate is not limited to such a condition, and the first candidate and the second candidate candidate characters are prioritized among the candidate characters, and candidates after the third candidate are selected. It may be configured to select only words that can be configured without using.

  FIG. 8 is a diagram illustrating a comparative example 46 between the recognition result character string and the word character string. In the comparative example 46, the upper row is a recognition result character string generated by the character string generator 19, and the lower row is a word character string generated by the word string generator 21. The lower word character string shown in FIG. 8 is a character string made up of words selected as the optimum word in the word character string 45 shown in FIG. In the comparative example 46, the recognition result character string and the word character string are separated by a rectangle divided by the number of characters constituting each word of the word character string.

  The replacement determination unit 22 compares the recognition result character string generated by the character string generation unit 19 and the word character string generated by the word string generation unit 21 by comparing one character at a time from the first character for each word. Determine whether. Whether or not the characters are different can be determined, for example, based on whether or not the character code of the character in the recognition result character string is different from the character code of the character in the word character string. When the character codes are different, it is determined that the characters are different, and when the character codes match, it is determined that the characters are not different. The character code can be realized, for example, by recording each character in the character recognition dictionary 14 and the word dictionary 24. In this embodiment, the characters are compared one by one from the first character, but may be compared one by one in order from the last character of the word to the first character.

  The replacement determination unit 22 determines that all characters constituting the word are replaced when there are no different characters for all the characters constituting the word, that is, when the characters are the same.

  When there are different characters, the replacement determination unit 22 further has a candidate in which the same character as the character determined to be different among the characters constituting the word is at a corresponding position other than the candidate characters constituting the recognition result character string. Determine if it is in a character. To be in the corresponding position means that the position from the beginning of the character in the word character string is the same as the position from the beginning of the recognition result character string. When the same character as the character determined to be different among the characters constituting the word is all in the candidate characters at the corresponding positions other than the candidate characters constituting the recognition result character string, the replacement determining unit 22 It is determined that all characters constituting the word are replaced.

  In other words, if all the characters that are the same as the characters that make up the word are among the candidate characters at the corresponding positions in the recognition result character string, the replacement determination unit 22 determines that all the characters that make up the word are replaced. .

  In the present embodiment, the replacement determination unit 22 searches for whether there is a matching candidate character among the candidate characters in this way, but when the word string generation unit 21 generates a word character string, each word is Information indicating whether or not a character that is a partially matching word and a character constituting the word is a character that does not exist in the candidate character is stored in the memory 13, and the replacement determination unit 22 stores the information stored in the memory 13. By referring to the information, it may be determined whether the candidate character is present.

  If the same character as any one of the characters constituting the word is not among the candidate characters at the corresponding position in the recognition result character string, that is, the replacement determination unit 22 recognizes the recognition result among the characters constituting the word. If there is no same character among the first to fifth candidate characters at the corresponding position of the character string, the first to fifth candidates at the corresponding position of the recognition result character string among the characters constituting the word For characters that do not have the same character among the candidate characters, it is further determined whether or not replacement is necessary based on the similarity. Hereinafter, among the characters constituting the word, the characters that do not have the same character among the first to fifth candidate characters at the corresponding positions in the recognition result character string are the candidate characters among the characters constituting the word. Characters that do not have the same character in

  Specifically, the replacement determination unit 22 determines the feature amount of the character that does not have the same character among the candidate characters, and the candidate character at the corresponding position that forms the recognition result character string. Based on the feature amount of one candidate character, the similarity between these feature amounts is calculated by Equation (3). The feature amount of the candidate character at the corresponding position constituting the recognition result character string is the feature amount extracted by the feature extraction unit 17 and stored in the memory 13. Then, when the calculated similarity is equal to or greater than a predetermined threshold, it is determined that a character that does not have the same character among candidate characters among characters constituting the word is determined to be replaced, and the calculated similarity is less than a predetermined threshold At this time, it is determined that a character that does not have the same character among the candidate characters among the characters constituting the word is not replaced. It is determined that a character having the same character among the candidate characters among the characters constituting the word is replaced. An arbitrary value can be set as the predetermined threshold. For example, when the maximum similarity of the degree of similarity calculated by the expression (3) = 70% of 128, “89” is a predetermined threshold value. That the similarity is equal to or higher than a predetermined threshold is an example of satisfying a predetermined comparison criterion.

  For example, the word “resident” in the word character string shown in FIG. 8 is a word corresponding to the character string “resident blue” in the recognition result character string shown in FIG. When compared, the third character is different. In this case, it is checked whether the third character “person” of the word “resident” in the word character string exists in the corresponding third character candidate character in the recognition result character string. Referring to the candidate character 43 shown in FIG. 6A, the third candidate character of the first candidate is the character “blue” different from the character “person”, and the candidate characters after the second candidate Does not have the word “person”.

  That is, in the example shown in FIG. 8, the character “blue” and the character “person” surrounded by the two-dot chain line 461 do not match, and the character “person” constituting the word is in the corresponding position. Since it does not exist in the candidate characters of the candidate to the fifth candidate, the replacement determination unit 22 features the feature amount of the character “person” recorded in the character recognition dictionary 14 and the feature of the corresponding candidate character constituting the recognition result character string. Compare the quantity and calculate the similarity. In the example shown in FIG. 8, the feature amount of the candidate character at the corresponding position constituting the recognition result character string is the character represented by the third character image in the character image 42 shown in FIG. This is the feature amount of the character extracted as the first candidate character “blue” by the matching unit 18. Since the feature amount extracted by the feature extraction unit 17 is stored in the memory 13, the replacement determination unit 22 refers to the memory 13 and extracts it by the feature extraction unit 17 without calculating the feature amount again. It is possible to know the feature amount.

  FIG. 9 is a diagram illustrating an example of character feature amounts of word character strings and feature amounts of candidate characters at corresponding positions. FIG. 9A shows feature data 39a indicating the feature amount of the candidate character “blue”. The candidate character “blue” is a character extracted by the matching unit 18 as the first candidate candidate character “blue” for the character represented by the third character image in the character image 42 shown in FIG. 4. FIG. 9B is feature data 32 a indicating the feature amount of the character “person” recorded in the character recognition dictionary 14. Although the actual calculation is omitted, the similarity between the feature amount of the candidate character “blue” and the feature amount of the character “person” recorded in the character recognition dictionary 14 is “108”. Here, the maximum value of the similarity is 128. In the example of the candidate character “blue” and the character “person” recorded in the character recognition dictionary 14, the replacement determination unit 22 has a calculated similarity “108” that is a similarity equal to or higher than a predetermined threshold = 89. Determine to replace.

  The replacement determination unit 22 determines whether or not to replace all words including characters that do not have the same character among the candidate characters among the characters constituting the word character string, based on the similarity. In the example shown in FIG. 8, among the characters constituting the word character string, characters that do not have the same character among the candidate characters include the character “preservation” in addition to the character “person”. Similarities are also calculated for the character “wa” and the character “K” enclosed by the dotted line 462, and it is determined whether or not to replace based on the calculated similarity.

  FIG. 9C shows feature data 39b indicating the feature amount of the candidate character “shake”. FIG. 9D shows feature data 32 b indicating the feature amount of the character “K” stored in the character recognition dictionary 14. Although the actual calculation is omitted, the degree of similarity between the feature amount of the candidate character “Ring” and the feature amount of the character “K” stored in the character recognition dictionary 14 is “110”. The replacement determination unit 22 replaces the similarity “110” between the feature amount of the candidate character “shake” and the feature amount of the character “K” recorded in the character recognition dictionary 14 because the predetermined threshold value is 89 or more. Is determined.

  In the present embodiment, the determination as to whether or not to replace is determined to be performed when the similarity is equal to or higher than a predetermined threshold, but the present invention is not limited to this. For example, the difference between the similarity calculated from the feature amount of the candidate character and the feature amount of the character recorded in the character recognition dictionary 14 and the similarity of the first candidate character of the target character is N or less. Sometimes, it may be determined to replace. N is an arbitrary positive integer.

  The replacement unit 23 replaces the character of the recognition result character string with the character of the word character string for the character determined to be replaced in the recognition result character string, and transmits the replaced recognition result character string to the output device 3 for output. . When it is determined not to replace all the characters in the recognition result character string, the recognition result character string is transmitted to the output device 3 and output without replacement.

  For example, in the example shown in FIG. 8, the characters “blue” and “person” surrounded by the two-dot chain line 461 and the characters “shake” and character “K” enclosed by the two-dot chain line 462 are replaced. Since it is determined to be replaced by the determination unit 22, the replacement unit 23 replaces the character “blue” in the recognition result character string with the character “person” in the word character string, and also replaces the character “vibration” in the recognition result character string. ”Is replaced with the character“ K ”in the word string.

  Characters other than the characters “blue” and the character “person” enclosed by the two-dot chain line 461 and the characters “wa” and the character “preservation” enclosed by the two-dot chain line 462 are the same characters, and thus are replaced. The replacement unit 23 replaces the character in the recognition result character string with the same character in the word character string. The replacement unit 23 sends the replaced recognition result character string to the output device 3 for output.

  That is, the recognition result character string “resident blue” is replaced with the word character string “resident”, and the recognition result character string “insurance” is replaced with the word character string “insurance”. Accordingly, the replacement unit 23 replaces and replaces the recognition result character string 44 shown in FIG. 6B, that is, “about the resident blue tremor medical system” with “about the resident insurance medical system”. The recognition result character string “about the resident's insurance medical system” is sent to the output device 3 and output.

  FIG. 10 is a flowchart illustrating a processing procedure of character recognition processing executed by the information processing device 1, the input device 2, and the output device 3. For example, when a document is set on the input device 2 and a document reading instruction is given by an operation unit (not shown) of the input device 2, the process proceeds to step A1.

  In step A1, the input device 2 reads the document as an image, and transmits image data representing the read image to the information processing device 1. In the information processing device 1, the control unit 11 receives the image data transmitted from the input device 2 by the input / output control unit 12, and sends the received image data to the character string row cutout unit 15. In step A2, the character string row cutout unit 15 cuts out the character string row image representing the character string from the image represented by the image data received from the control unit 11 for each row. In step A3, the character cutout unit 16 cuts out a character image that is a rectangular image representing characters one by one from the character string row image cut out by the character string row cutout unit 15.

  In step A <b> 4, the feature extraction unit 17 extracts a feature amount obtained by quantifying the character feature represented by the character image cut out by the character cutout unit 16. In step A5, the matching unit 18 performs matching with the character recognition dictionary 14. Specifically, the matching unit 18 compares the feature amount extracted by the feature extraction unit 17 with the feature amount of all the characters recorded in the character recognition dictionary 14 to determine the character and character represented by the character image. The degree of similarity with the characters recorded in the recognition dictionary 14 is calculated, and at least one candidate character for the character represented by the character image, for example, the first to fifth candidate characters is calculated based on the calculated similarity. Extract.

  In step A6, the character string generation unit 19 generates a recognition result character string in which candidate characters of the first candidate are arranged in order from the left among the candidate characters extracted by the matching unit 18. In step A7, the post-processing unit 20 performs the first post-processing and ends the character recognition processing.

  Step A2 shown in FIG. 10 corresponds to a character string row cutout process, step A3 shown in FIG. 10 corresponds to a character cutout process, step A4 shown in FIG. 10 corresponds to a feature extraction process, Step A5 shown in FIG. 10 corresponds to a selection process, and step A6 shown in FIG. 10 corresponds to a character string generation process.

  FIG. 11 is a flowchart showing the processing procedure of the first post-processing executed by the post-processing unit 20. When step A7 shown in FIG. 10 is executed, the process proceeds to step B1.

  In step B1, the word string generation unit 21 generates a word character string from the recognition result character string generated by the character string generation unit 19. Specifically, a character string generated by combining a plurality of candidate characters of each character constituting the recognition result character string generated by the character string generation unit 19 and a word recorded in the word dictionary 24 Word candidates are generated by collation, an optimal word is selected from the generated word candidates, and a word character string composed of the selected optimal word is generated.

  In step B <b> 2, the replacement determination unit 22 determines whether a recognition result character string exists for all characters in the character string constituting the word for each word constituting the word character string. That is, when step B2 is executed for the first time, for all characters constituting the first word constituting the word character string, when step B2 is executed after the second time, the word is processed after the previously processed word. It is determined whether or not there is the same character among the first to fifth candidates at the corresponding position in the recognition result character string for all the constituent characters. Hereinafter, the word which is the target of processing among the words constituting the word character string in step B2 is referred to as “target word”.

  For all characters that constitute a word in the word character string, if there is the same character among the first to fifth candidates at the corresponding position in the recognition result character string, all characters in the character string that constitute the word In step S3, it is determined that the recognition result character string exists, and the process proceeds to step B3. If any of the characters constituting the word of the word character string does not have the same character among the first to fifth candidates at the corresponding position in the recognition result character string, all of the characters in the character string constituting the word In the character, it is determined that the recognition result character string does not exist, and the process proceeds to Step B5.

  When the process proceeds to step B3, the replacement determination unit 22 determines to replace the determined character in the recognition result character string. When the process proceeds to step B5, if there is a character that is determined to have the same character among the first to fifth candidates at the corresponding position in the recognition result character string for the characters constituting the word, the recognition result It is determined that the character that has been determined in the character string is to be replaced.

  In step B3, the replacement unit 23 outputs a word character for the target character. The target character in step B3 is a character determined to be replaced by the replacement unit 23. A word character is a character which comprises a word character string. The output at step B3 is to be stored in the memory 13. That is, the replacement unit 23 replaces the character determined to be replaced by the replacement unit 23 among the target words with the character of the word character string, and stores the replaced word in the memory 13. At this time, if there is a character determined not to be replaced in the same word, the character determined not to be replaced is not replaced.

  In step B4, it is determined whether or not processing has been performed for all words constituting the word character string. The process in step B4 is the process in steps B2, B3, B5 to B7. When the processing has been performed for all the words constituting the word character string, the replacement unit 23 transmits the recognition result character string composed of the words stored in the memory 13 to the output device 3 and outputs the recognition result character string. End post-processing. When processing has not been performed for all the words constituting the word character string, the process returns to step B2.

  In step B5, the replacement determination unit 22 compares the character feature value of the word dictionary with the feature value of the recognized character. The characters in the word dictionary are characters that have been determined in step B2 that no candidate character has the same character among the characters constituting the word in the word character string. The recognized character is a candidate character at a corresponding position among candidate characters constituting the recognition result character string, that is, a candidate character of the first candidate. That is, in step B5, the replacement determination unit 22 sets the first feature character in the character that does not have the same character among the candidate characters among the characters constituting the word of the word character string and the first position at the corresponding position of the recognition result character string. The feature amount of the candidate character candidate is compared. For the purpose of comparison, the replacement determining unit 22 compares the feature amount of characters that do not have the same character among the candidate characters among the characters constituting the words of the word character string, and the first candidate at the corresponding position in the recognition result character string. The similarity with the feature amount of the candidate character is calculated.

  In step B <b> 6, the replacement determination unit 22 determines whether similarity equal to or greater than a predetermined threshold is obtained from the comparison of feature amounts. When the similarity calculated in step B5 is equal to or greater than a predetermined threshold, it is determined from the feature amount comparison that similarity equal to or greater than the predetermined threshold is obtained, and the process proceeds to step B3. When the degree of similarity calculated in step B5 is less than a predetermined threshold, it is determined from the feature amount comparison that no similarity equal to or higher than the predetermined threshold has been obtained, and the process proceeds to step B7.

  When the process proceeds to step B3, the replacement determination unit 22 determines to replace a character determined to have a similarity equal to or greater than a predetermined threshold in the recognition result character string. When the process proceeds to step B7, it is determined that the character that is determined to have no similarity equal to or greater than a predetermined threshold in the recognition result character string is not replaced.

  In step B7, the replacement unit 23 outputs a recognized character for the target character, and proceeds to step B4. The target character in step B7 is a character determined not to be replaced by the replacement unit 23. The output at step B7 is to be stored in the memory 13. That is, the replacement unit 23 stores the word to be recognized in the memory 13 without replacing the characters determined to be not replaced by the replacement unit 23 among the target words with the characters of the word character string. At this time, if there is a character determined to be replaced in the same word, the character in the recognition result character string is replaced with the character in the word character string for the character determined to be replaced.

  Step B1 shown in FIG. 11 corresponds to a word string generation process, steps B2, B4 to B6 shown in FIG. 11 correspond to a replacement determination process, and steps B3 and B7 shown in FIG. It corresponds to.

  In the present embodiment, the replacement unit 23 temporarily stores words composed of replaced characters and non-replaced characters in the memory 13, and after all the words have been processed, the words stored in the memory 13. Although the recognition result character string is constructed from the above and transmitted to the output device 3 and outputted, it is composed of the replaced character and the non-replaced character in step B3 or step B7 without being stored in the memory 13. It is good also as a structure which transmits and outputs the word to the output device 3 for every word.

  Next, the information processing apparatus 1a which is the 2nd Embodiment of this invention is demonstrated. The information processing method according to the present invention is executed by the information processing apparatus 1a. The configuration of the information processing apparatus 1a is the same as the configuration of the information processing apparatus 1 shown in FIG. 1 except for the post-processing unit 20a, and only different parts will be described below in order to avoid duplication. The processing executed by the post-processing unit 20a is hereinafter referred to as second post-processing. The post-processing unit 20a is the same as the configuration of the post-processing unit 20 shown in FIG. 1 except for the replacement determination unit 22a, and only different parts will be described below in order to avoid duplication.

  The replacement determination unit 22a first stores in the memory 13 whether or not similar words have been extracted for the words constituting the word character string generated by the word string generation unit 21. When a word character string is generated by the word string generator 21, when a plurality of word candidates are generated, the remaining words other than the word candidate having the highest similarity among the plurality of word candidates are generated. It is a candidate. This process is a process that was not performed by the replacement determination unit 22 of the first embodiment.

  Next, similarly to the replacement determination unit 22 of the first embodiment, the replacement determination unit 22a detects the candidate character that has the same character as the character constituting the word of the word character string at the corresponding position in the recognition result character string. Determine if it is inside. When the same characters as the characters constituting the word in the word character string are all in the candidate characters at the corresponding positions in the recognition result character string, the replacement determination unit 22 replaces all the characters constituting the word. judge.

  When there is no same character among the candidate characters at the corresponding position in the recognition result character string among the characters constituting the word, the replacement determination unit 22a selects the word as in the replacement determination unit 22 of the first embodiment. From the feature amount of characters that do not have the same character among the candidate characters and the corresponding candidate characters constituting the recognition result character string, that is, the feature amount of the candidate character of the first candidate, among the constituent characters, according to equation (3) The similarity between these feature quantities is calculated. Then, when the calculated similarity is equal to or greater than a predetermined threshold, it is determined to replace a character in the recognition result character string at a position corresponding to a character that does not have the same character among candidate characters among characters constituting the word. .

When the calculated similarity is less than a predetermined threshold, the replacement determination unit 22a is different from the replacement determination unit 22 of the first embodiment, and it is further determined that the calculated similarity is less than a predetermined threshold. It is determined with reference to the memory 13 whether or not there are similar words in the words including characters. When there is no similar word in a word including a character determined to have a calculated similarity less than a predetermined threshold, a word including a character determined to have a calculated similarity less than a predetermined threshold has a similarity It is determined that the certainty level is high as in the case of high, and it is determined to replace. When there is a similar word in a word including a character that has been determined that the calculated similarity is less than a predetermined threshold, it is determined not to be replaced.
Hereinafter, as a specific example, the processing of the information processing apparatus 1a according to the second embodiment will be described using the character string row image 412 illustrated in FIG.

  FIG. 12 is a diagram illustrating an example of the character image 52 cut out by the character cutout unit 16. The character cutout unit 16 cuts out characters one by one as a rectangular image from the character string row image cut out by the character string row cutout unit 15. In the example of the character image 52 shown in FIG. 12, a character image of 22 characters cut out from the character string row image 412 shown in FIG. 3 is shown. The first character image 521 is a character image obtained by cutting out the character “Kyo”, and “power”, “,”, “reason”, “solution”, “o”, “gain”, “ ``, '' `` Ta '', `` me '', `` ni '', `` thing '', `` before '', `` ni '', `` theory '', `` akira '', `` meeting '', `` to '', `` row '', `` tsu '' , “Ta” and “.” Are shown as character images.

  FIG. 13 is a diagram illustrating an example of the candidate character 53 and the recognition result character string 54. FIG. 13A shows an example of the candidate character 53 extracted by the matching unit 18. The candidate character 53 is a candidate character extracted by the matching unit 18 as a character represented by the character image 52. In the example of the candidate character 53 shown in FIG. 13A, “Kyo”, “Power”, “,”, “Science”, “Solution”, “ "," "Gain", "ru", "ta", "me", "ni", "thing", "previous", "ni", "theory", "akira", "meeting", "to" , “Line”, “tsu”, “ta” and “.” Are shown. As for the first to fifth candidates, for example, for the correct character “Kyo”, the first candidate “Kyo 121”, the second candidate “Waki 118”, the third candidate “Suri 110”, and the fourth candidate “Enjoyment 108”. And the fifth candidate “Western 104” is shown.

  FIG. 13B shows an example of the recognition result character string 54 generated by the character string generation unit 19. The character string generation unit 19 generates a recognition result character string 54 in which the first candidate characters are arranged in order from the left among the candidate characters indicated by the candidate character 53 extracted by the matching unit 18. The character string generation unit 19 stores candidate characters 43 in the memory 13 for the post-processing unit 20a. In the example shown in FIG. 13B, the recognition result character string 54 is a character string “A briefing was held between days to obtain cooperation and buried”. When the processing by the post-processing unit 20a is not performed, the recognition result character string 54 is output as a recognition result. In the present embodiment, thereafter, the second post-processing is executed by the post-processing unit 20a.

  In the second post-processing, first, the word string generation unit 21 stores each candidate character in the word dictionary 24 in order from the first character of the recognition result character string, as in the first embodiment. A character string that completely matches or partially matches a word is generated as a word candidate. An example of the word character string generated in this way is shown in FIG.

  In the second embodiment, unlike the first embodiment, the word string generation unit 21 stores in the memory 13 whether or not a plurality of word candidates are generated for the post-processing unit 20a. . Specifically, when the word candidate included in the generated word character string is one of the word candidates from which a plurality of word candidates are generated, the fact that similar words have been extracted is stored in the memory 13. If the word candidate included in the generated word character string is a word candidate for which only one word candidate has been generated, the memory 13 stores that similar words are not extracted. The similar word is a word candidate that is not included in the generated word character string among the plurality of generated word candidates.

  FIG. 14 is a diagram illustrating an example of the word character string 55 generated by the word string generation unit 21. In the example of the word character string 55 shown in FIG. 14, an optimal word such as a word 551 and a plurality of generated word candidates such as word candidates 552 and 553 are mixed. Two word candidates “Get” and “Get” are generated for the word candidate 552, and two word candidates “advance” and “specialty” are generated for the word candidate 553.

  Next, the word string generation unit 21 selects an optimal word for a word character string 55 having a plurality of word candidates, and generates a word character string including only the optimal word candidates. In the example shown in FIG. 14, since there are a plurality of word candidates in the word candidates 552 and 553, the word string generation unit 21 assigns priorities under the same three conditions as in the first embodiment, Select a word.

  For example, in the example shown in FIG. 14, the word candidates 552 “Get” and “Get” both satisfy the condition 1, so the optimum word is selected according to the condition 3. Of “Get” and “Get”, “Get” has a higher similarity, so “Get” is selected. For “advanced” and “specialized”, “advanced” satisfies condition 1, but “advanced” does not satisfy condition 1 but satisfies condition 2; therefore, “advanced” that satisfies a high priority condition is selected. .

  FIG. 15 is a diagram illustrating a comparative example 56 of the recognition result character string and the word character string. In the comparative example 56, the upper row is a recognition result character string generated by the character string generator 19, and the lower row is a word character string generated by the word string generator 21. The lower word character string shown in FIG. 15 is a character string made up of words selected as optimum words in the word character string 55 shown in FIG. In the comparative example 56, the recognition result character string and the word character string are separated by a rectangle divided by the number of characters constituting each word of the word character string.

  Subsequently, the replacement determination unit 22a compares the recognition result character string generated by the character string generation unit 19 and the word character string generated by the word string generation unit 21 one by one from the first character for each word, Determine whether to replace. The determination as to whether or not to replace is the same as in the first embodiment except for the following.

  In the first embodiment, when the calculated similarity is less than a predetermined threshold, the replacement determination unit 22 determines that a character that does not have the same character among the candidate characters among the characters constituting the word is not replaced. However, in the second embodiment, the replacement determination unit 22a determines whether a similar word has been extracted when the calculated similarity is less than a predetermined threshold. If the replacement determination unit 22a refers to the memory 13 and does not store that the similar word is extracted in the memory 13, the replacement determination unit 22a determines that the replacement is performed and stores that the similar word is extracted in the memory 13. It is determined that it will not be replaced.

  For example, in the example shown in FIG. 15, determination as to whether or not the replacement determination unit 22a replaces will be described. When the characters are compared in order from the front, the characters in the fourth character and the fifth character, that is, the portion surrounded by the two-dot chain line 561 are different. That is, the recognition result character string is “buried”, and the word character string is “understanding”. In this embodiment, the recognition result character string and the word character string are compared in order from the front, but may be configured to compare in order from the back.

  The replacement determination unit 22a first checks whether or not the fourth character “Ri” in the word character string exists in the candidate character at the corresponding position in the recognition result character string. In the candidate character 53 shown in FIG. 13A, there is a character “RI” in the candidate character of the second candidate of the fourth character. Therefore, since the character “RI” is in the candidate character, no further processing is performed on the character “RI”. Next, it is checked whether or not the fifth character “solution” exists in the candidate character at the corresponding position in the recognition result character string. In the candidate character 53 shown in FIG. 13A, the candidate character of the first candidate of the fifth character is “斛”, and the character “solution” does not exist after the candidate character of the second candidate.

  Since the character “solution” constituting the word “understanding” is not included in the candidate characters of the recognition result character string, the replacement determination unit 22a next determines whether or not to replace based on the similarity. At this time, if all characters of the word “understanding” are included in the candidate characters of the recognition result character string, it is determined to be replaced.

  Since the character “solution” in the word character string does not exist in the candidate character at the corresponding position in the recognition result character string, the replacement determination unit 22a has the feature of the character “solution” recorded in the character recognition dictionary 14. The degree of similarity is extracted by comparing the amount with the feature amount extracted from the character image of the fifth character in the character image 52 shown in FIG. The feature amount extracted from the character image of the fifth character in the character image 52 shown in FIG. 12 is stored in the memory 13.

  FIG. 16 is a diagram illustrating an example of character feature amounts of word character strings and feature amounts of candidate characters at corresponding positions. FIG. 16A shows feature data 39c indicating the feature amount of the candidate character “斛”. The candidate character “斛” is a character extracted by the matching unit 18 as the first candidate candidate character “斛” for the character represented by the fifth character image in the character image 52 shown in FIG. 12. FIG. 16B is feature data 32 c indicating the feature amount of the character “solution” recorded in the character recognition dictionary 14. Although the actual calculation by Expression (3) is omitted, the similarity between the feature amount of the candidate character “斛” and the feature amount of the character “solution” recorded in the character recognition dictionary 14 is “112”. Here, the maximum value of the similarity is 128.

  The predetermined threshold value can be set as an arbitrary value. For example, when the predetermined threshold value is less than 5 in similarity with the first candidate candidate character, the first candidate of the fifth character The candidate character is “斛”, and the similarity of the character “斛” is indicated as “119” in FIG. 13A. Therefore, the predetermined threshold is “114” obtained by subtracting “5” from “119”. "become. The similarity “112” between the character image corresponding to the fifth character and the character “solution” is less than a predetermined threshold = 114, and the presence / absence of a similar word is determined. If the degree of similarity is equal to or greater than a predetermined threshold, it is determined to replace.

  In determining whether or not there is a similar word, the replacement determining unit 22a refers to the memory 13 to determine whether or not a similar word has been extracted in addition to the word “understanding”. If it is not stored in the memory 13 that the similar word has been extracted, the word “understanding” has a high degree of certainty with respect to the character string “buried”, and the degree of similarity is equal to or higher than a predetermined threshold. It is determined that the same processing is performed, and replacement is determined. If it is stored in the memory 13 that similar words have been extracted, it is determined that the word “understanding” is not highly reliable with respect to the character string “buried”, and the degree of similarity is less than a predetermined threshold. It is determined that the same processing is performed as in the case of, and it is determined not to be replaced.

  In the second embodiment, if the fact that similar words have been extracted is not stored in the memory 13, the certainty factor is unconditionally high. However, the certainty factor is not unconditionally high. For example, the predetermined threshold value for determining the similarity between the character “solution” and the character “solution” may be reset to 50% of the maximum similarity = 128.

  In the second post-processing, finally, the replacement unit 23 replaces the characters in the recognition result character string with the characters in the word character string for the characters determined to be replaced in the recognition result character string, as in the first embodiment. And the replaced recognition result character string is transmitted to the output device 3 for output. When it is determined not to replace all the characters in the recognition result character string, the recognition result character string is transmitted to the output device 3 and output without replacement.

  Since the word “understanding” in the word character string shown in FIG. 15 is not stored in the memory 13 that the similar word is extracted, it is called “buried” in the recognition result character string shown in FIG. For the character string, it is determined that the word “understanding” has a high certainty factor, and is replaced in the same manner as when the similarity is equal to or higher than a predetermined threshold. Therefore, the replacement unit 23 replaces the character string “buried” in the recognition result character string with the word “understanding” in the word character string, and transmits it to the output device 3 for output. That is, by outputting the word “understanding” in the word character string, the word erroneously recognized as “buried” can be correctly corrected, that is, corrected.

  In the example shown in FIG. 15, in addition to “buried” of “buried” in the recognition result character string and “reason” of “understanding” in the word character string, “ameni” in the recognition result character string “A” and “Ta” of “For” in the word string, “Between” of “Things” in the recognition result string, and “Previous” of “Pre” in the word string Since the characters are different from each other, it is determined whether or not replacement is necessary based on the similarity.

  Although explanations of these comparison results are omitted, in the second embodiment, the characters in the recognition result character string are different from the characters in the word character string, and the similarity is less than a predetermined threshold. However, if the certainty of the words constituting the word character string is high, a correct recognition result can be obtained by unconditionally replacing the characters in the recognition result character string with the characters in the word character string. Alternatively, even if the character in the recognition result character string is different from the character in the word character string, and the similarity is less than a predetermined threshold, the predetermined threshold is reset to a lower value and the similarity is set. When the corrected threshold value is equal to or higher than the predetermined threshold, it is possible to obtain a correct recognition result by replacing characters in the recognition result character string with characters in the word character string, assuming that the certainty of the words constituting the word character string is high. it can.

  FIG. 17 is a flowchart showing the procedure of the second post-processing executed by the post-processing unit 20a. The character recognition processing executed by the information processing device 1a, the input device 2 and the output device 3 is the same as the character recognition processing executed by the information processing device 1, the input device 2 and the output device 3, and will not be described in order to avoid duplication. Omitted. In the second embodiment, when step A7 shown in FIG. 10 is executed, the process proceeds to step C1. Steps C1, C3 to C6 and C9 are the same as steps B1 to B5 and B7 shown in FIG.

  In Step C2, the word string generation unit 21 stores, in the memory 13, for each word constituting the generated word character string, whether or not a similar word has been extracted from the words constituting the generated word character string. That is, when the word included in the generated word character string is one word candidate among the plurality of generated word candidates, the fact that a similar word has been extracted is stored in the memory 13, and the generated word When the word included in the character string is a candidate for only one generated word, it is stored in the memory 13 that a similar word is not extracted.

  In step C7, the replacement determination unit 22a determines whether similarity equal to or higher than a predetermined threshold is obtained from the comparison of feature amounts. When the similarity calculated in step C6 is equal to or greater than a predetermined threshold, it is determined from the feature amount comparison that similarity equal to or greater than the predetermined threshold is obtained, and the process proceeds to step C4. When the degree of similarity calculated in step C6 is less than a predetermined threshold, it is determined from the feature amount comparison that no similarity equal to or higher than the predetermined threshold has been obtained, and the process proceeds to step C8.

  Although the predetermined threshold value can be set as an arbitrary value, in the present embodiment, for example, the predetermined threshold value is a value obtained by subtracting “5” from the similarity with the candidate character of the first candidate. In the example shown in FIG. 16A, the candidate character of the first candidate for the fifth character is “斛”, and the similarity of the character “斛” is shown as “119” in FIG. Therefore, the predetermined threshold value is “114”.

  In step C8, the replacement determination unit 22a determines whether or not a similar word has been extracted in step C2 for the target word. If it is stored in the memory 13 that similar words have been extracted for the word to be processed, that is, the word to be processed in step C3 in the word character string, the process proceeds to step C9. If it is stored in the memory 13 that no similar word has been extracted, the process proceeds to step C4. When the process proceeds to step C4, the replacement determination unit 22a determines to replace a character determined to have a similarity equal to or greater than a predetermined threshold in the recognition result character string. When the process proceeds to step C9, it is determined that the character that is determined not to have a similarity equal to or higher than a predetermined threshold in the recognition result character string is not replaced.

  Steps C1 and C2 shown in FIG. 17 correspond to a word string generation process, steps C3 and C5 to C8 shown in FIG. 17 correspond to a replacement determination process, and steps C4 and C9 shown in FIG. This corresponds to the replacement process.

  Next, the information processing apparatus 1b which is the 3rd Embodiment of this invention is demonstrated. The information processing method according to the present invention is executed by the information processing apparatus 1b. The configuration of the information processing apparatus 1b is the same as the configuration of the information processing apparatus 1 shown in FIG. 1 except for the post-processing unit 20b, and only different parts will be described below in order to avoid duplication. The processing executed by the post-processing unit 20b is hereinafter referred to as third post-processing. The post-processing unit 20b is the same as the configuration of the post-processing unit 20 shown in FIG. 1 except for the replacement determination unit 22b, and only different parts will be described below in order to avoid duplication.

  The replacement determination unit 22b, first, in the same way as the replacement determination unit 22 of the first embodiment, the same character as the character constituting the word of the word character string is among the candidate characters at the corresponding position of the recognition result character string. It is determined whether or not. When the same characters as the characters constituting the word in the word character string are all in the candidate characters at the corresponding positions in the recognition result character string, the replacement determination unit 22 replaces all the characters constituting the word. judge.

  When there is no same character among the candidate characters at the corresponding position in the recognition result character string among the characters constituting the word, the replacement determination unit 22b determines the word as in the replacement determination unit 22 of the first embodiment. From the feature amount of characters that do not have the same character among the candidate characters and the corresponding candidate characters constituting the recognition result character string, that is, the feature amount of the candidate character of the first candidate, among the constituent characters, according to equation (3) The similarity between these feature quantities is calculated. Then, when the calculated similarity is equal to or greater than a predetermined threshold, it is determined to replace a character in the recognition result character string at a position corresponding to a character that does not have the same character among candidate characters among characters constituting the word. .

  When the calculated similarity is less than a predetermined threshold, the replacement determination unit 22b is different from the replacement determination unit 22 of the first embodiment, and whether or not the number of characters of the target word is equal to or greater than a predetermined reference character number. Determine. The target word is a word including a character whose similarity is determined among words constituting the word character string. If it is not equal to or more than the predetermined standard number of characters, it is determined not to be replaced.

  When the number of characters is equal to or greater than the predetermined reference character number, the predetermined threshold value is reset to a lower value, and when the similarity is equal to or higher than the predetermined threshold value that is reset, it is determined to be replaced. This is because, in the case of a character string having a long number of characters, the word selected in the word character string can be matched with the word recorded in the word dictionary 24 completely or partially. For words that are considered to be high and have a predetermined number of reference characters or more, a predetermined threshold value used for determining the similarity is set to a lower value. When the similarity is less than a preset threshold that has been reset, it is determined that the replacement is not performed.

  FIG. 18 is a diagram illustrating an example of an image 60 represented by image data received from the input device 2. An image 60 represented by image data received from the input device 2 is an image obtained by reading a document on which a character string row is written by the input device 2. In the image 60, a character string line is displayed. In the example of the image 60 shown in FIG. 13, one character string row “Need to submit after obtaining the garage certificate” is displayed.

  Hereinafter, as a specific example of the third embodiment, the processing of the information processing apparatus 1b will be described using the image 60 illustrated in FIG.

  FIG. 19 is a diagram illustrating an example of the character string row image 61 cut out by the character string row cutout unit 15. The character string row cutout unit 15 receives the image data received from the input device 2 by the control unit 11 from the input device 2 by the input / output control unit 12, and generates a character string row image representing a character string from the image represented by the received image data. Cut out line by line. In the example of the character string row image 61 illustrated in FIG. 19, one character string row image of the character string row image 611 is illustrated. The character string row image 611 is an image in which a character string “It is necessary to submit after obtaining the garage certificate” is displayed.

  FIG. 20 is a diagram illustrating an example of a character image 62 cut out by the character cutout unit 16. The character cutout unit 16 cuts out characters one by one as a rectangular image from the character string row image cut out by the character string row cutout unit 15. In the example of the character image 62 shown in FIG. 20, a character image of 19 characters is shown. The first character image 621 is a character image obtained by cutting out the character “car”. The second and subsequent characters are “warehouse”, “proof”, “light”, “no”, “take”, “gain”, “ After, “,”, “proposed”, “out”, “su”, “ru”, “necessary”, “necessary”, “ga”, “a”, “ru” and “.” A character image is shown.

  FIG. 21 is a diagram illustrating an example of the candidate character 63 and the recognition result character string 64. FIG. 21A shows an example of the candidate character 63 extracted by the matching unit 18. The candidate character 63 is a candidate character extracted by the matching unit 18 as a character represented by the character image 62. In the example of the candidate character 53 shown in FIG. 21A, the input character at the top, that is, the correct character of the candidate character is “car”, “storage”, “proof”, “light”, “no”, “ `` Take '', `` gain '', `` after '', ``, '', `` suggest '', `` out '', `` su '', `` ru '', `` must '', `` must '', `` ga '', `` a '', `` ru '' And “.” Are shown. For the first to fifth candidates, for example, for the correct character “car”, the first candidate “Matter 123”, the second candidate “Exclusive 120”, the third candidate “Act 119”, and the fourth candidate “Required 118”. ”And the fifth candidate is blank.

  FIG. 21B shows an example of the recognition result character string 64 generated by the character string generation unit 19. The character string generation unit 19 generates a recognition result character string 64 in which the first candidate characters are arranged in order from the left among the candidate characters indicated by the candidate character 63 extracted by the matching unit 18. The character string generation unit 19 stores the candidate character 63 in the memory 13 for the post-processing unit 20b. In the example shown in FIG. 21B, a character string “It is necessary to submit after obtaining the case certificate” is shown as the recognition result character string 64. When the processing by the post-processing unit 20b is not performed, the recognition result character string 64 is output as a recognition result. In the present embodiment, thereafter, the third post-processing is executed by the post-processing unit 20b.

  In the third post-processing, first, the word string generation unit 21 records each candidate character in the word dictionary 24 in order from the first character of the recognition result character string, as in the first embodiment. A character string that completely matches or partially matches a word is generated as a word candidate. An example of the word character string generated in this way is shown in FIG.

  FIG. 22 is a diagram illustrating an example of the word character string 65 generated by the word string generation unit 21. In the example of the word character string 65 shown in FIG. 22, a word character string made up of an optimum word, for example, the word 651 is shown. The word character string 65 does not show a plurality of word candidates. The word string generation unit 21 refers to the word dictionary 24 as “garage proof” for the character string “case certificate” of the first to fourth characters in the recognition result character string 64 shown in FIG. If the word is recorded, this “garage proof” is determined to be a partially matching character string in which only the first characters “car” and “thing” differ from the “case proof”. The “garage certificate” is extracted as a word corresponding to the “case certificate”. Other words such as “accident” or “garage” can be extracted, but none of them are exact matches, and the string length is shorter than “garage proof”. Only the longest “garage certificate” is extracted. The “garage proof” is extracted as a partially matched word.

  FIG. 23 is a diagram illustrating a comparative example 66 of the recognition result character string and the word character string. In the comparative example 66, the upper row is a recognition result character string generated by the character string generator 19, and the lower row is a word character string generated by the word string generator 21. The lower word character string shown in FIG. 23 is a character string made up of words selected as optimum words in the word character string 65 shown in FIG. In the comparative example 66, the recognition result character string and the word character string are separated by a rectangle divided by the number of characters constituting each word of the word character string.

  Subsequently, the replacement determination unit 22b compares the recognition result character string generated by the character string generation unit 19 and the word character string generated by the word string generation unit 21 one by one from the first character for each word, Determine whether to replace. The determination as to whether or not to replace is the same as in the first embodiment except for the following.

  In the first embodiment, when the calculated similarity is less than a predetermined threshold, the replacement determination unit 22 determines that a character that does not have the same character among the candidate characters among the characters constituting the word is not replaced. However, in the third embodiment, when the calculated similarity is less than a predetermined threshold, the replacement determination unit 22b further determines whether or not the number of characters of the target word is greater than or equal to a predetermined reference character number. The replacement determination unit 22b determines that the replacement is not performed when the number of characters of the target word is not greater than or equal to a predetermined reference number of characters. If the number of characters of the target word is equal to or greater than the predetermined reference character number, the predetermined threshold is reset to a lower value, and when the similarity is equal to or higher than the predetermined threshold that has been reset, it is determined to replace and the similarity is set. If it is not equal to or more than the predetermined threshold value, it is determined not to be replaced.

  For example, with respect to the example shown in FIG. 23, the determination as to whether or not to replace by the replacement determination unit 22b will be described. The word “garage proof” of the word character string shown in FIG. 23 corresponds to the word “case proof” of the recognition result character string shown in FIG. 23, and the characters are compared one by one in order from the front. The first character is different. The replacement determination unit 22b checks whether the character “car” obtained from the word dictionary 24 exists in the candidate character at the corresponding position in the recognition result character string. In this embodiment, the recognition result character string and the word character string are compared in order from the front, but may be configured to compare in order from the back.

  In the candidate character 63 shown in FIG. 21A, the candidate character “thing” of the first candidate of the first character of the recognition result character string is shown, and “car” after the candidate character below the second candidate. There is no such character. Since the “car” of the word “garage proof” in the word character string is not included in the candidate characters of the recognition result character string, that is, does not exist, the replacement determination unit 22b determines whether or not to replace based on the similarity. At this time, if all characters of the word “garage certificate” in the word character string are included in the candidate characters of the recognition result character string, the replacement determination unit 22b determines to replace.

  When there is a different character between the character in the recognition result character string and the character in the word character string, the replacement determination unit 22b determines the character feature amount in the recognition result character string and the character feature amount in the word character string. Are compared to extract the degree of similarity. For example, since the character “thing” and the character “car” enclosed by the two-dot chain line 661 shown in FIG. 23 are different, the feature amount of the character “car” recorded in the character recognition dictionary 14 and the recognition result character string The similarity is extracted by comparing with the feature amount extracted from the character image of the first character. The feature amount extracted from the character image of the first character of the recognition result character string, that is, the feature amount extracted from the character image 621 of the first character in the character image 62 shown in FIG. Has been.

  FIG. 24 is a diagram illustrating an example of character feature amounts of word character strings and feature amounts of candidate characters at corresponding positions. FIG. 24A shows feature data 39d indicating the feature amount of the candidate character “thing”. The candidate character “thing” is a character extracted by the matching unit 18 as the first candidate candidate character “thing” for the character represented by the first character image 621 in the character image 62 shown in FIG. . FIG. 24B shows feature data 32 d indicating the feature amount of the character “car” recorded in the character recognition dictionary 14. Although the actual calculation by Expression (3) is omitted, the similarity between the feature amount of the candidate character “thing” and the feature amount of the character “car” recorded in the character recognition dictionary 14 is “116”. Here, the maximum value of the similarity is 128.

  The predetermined threshold can be set as an arbitrary value. For example, when the predetermined threshold is a difference in similarity between the first candidate candidate character and the difference is within 5, the first candidate of the first character The candidate character is “Thing”, and the similarity of the character “Thing” is indicated as “123” in FIG. 21A. Therefore, the predetermined threshold is “118” obtained by subtracting “5” from “123”. "become. The similarity “116” between the character image corresponding to the first character and the character “car” is less than a predetermined threshold “118”, and the number of characters is determined. If the degree of similarity is equal to or greater than a predetermined threshold, it is determined to replace.

  In the determination of the number of characters, the replacement determination unit 22b determines whether or not the number of characters of the target word is greater than or equal to a predetermined reference number of characters. If the number of characters of the target word is not equal to or greater than the predetermined number of characters, the replacement determination unit 22b determines not to replace. In the case of a character string having a long number of characters, the fact that it can be matched completely or partially with a word recorded in the word dictionary 24 is considered to have a high probability that the extracted word is optimal, and is determined in advance. For words that exceed the number of reference characters, the predetermined threshold value used for similarity determination is reset to a lower value.

  The predetermined reference character number is, for example, “4”, and the predetermined threshold value that is reset is a value obtained by subtracting “10” from the similarity of the candidate character of the first candidate, or 80% of the maximum similarity = 128, For example, 102. The replacement determination unit 22b determines that the replacement is performed when the similarity is equal to or higher than a predetermined threshold that has been reset, and determines that the replacement is not performed when the similarity is less than the predetermined threshold that has been reset.

  In the example of the first character shown in FIG. 23, when the preset threshold value that has been reset is a value obtained by subtracting “10” from the similarity of the candidate character of the first candidate, the preset threshold value that has been reset Is “113”, and the similarity “116” between the character image corresponding to the first character and the character “car” is equal to or higher than the preset threshold value, and the replacement determination unit 22b determines to replace To do.

  In the third post-processing, finally, the replacement unit 23 replaces the characters of the recognition result character string with the characters of the word character string for the characters determined to be replaced among the recognition result character strings, as in the first embodiment. And the replaced recognition result character string is transmitted to the output device 3 for output. When it is determined not to replace all the characters in the recognition result character string, the recognition result character string is transmitted to the output device 3 and output without replacement.

  Since it is determined that the “thing” of the word “case certificate” in the recognition result character string illustrated in FIG. 23 is to be replaced, the replacement unit 23 performs “thing” of the word “case certificate” of the recognition result character string. Is replaced with "car" in the word character string "garage certificate", and the replaced word "garage certificate" is transmitted to the output device 3 for output. That is, the word “thing” of the misrecognized word “case certificate” of the words constituting the recognition result character string is collected by the “car” of the word string “garage certificate”, that is, corrected and output. Can do. Therefore, the information processing apparatus 1b can output the character string “It is necessary to submit after obtaining the garage certificate” as the recognition result.

  FIG. 25 is a flowchart illustrating a processing procedure of third post-processing executed by the post-processing unit 20b. The character recognition processing executed by the information processing device 1b, the input device 2 and the output device 3 is the same as the character recognition processing executed by the information processing device 1, the input device 2 and the output device 3, and will not be described in order to avoid duplication. Omitted. In the third embodiment, when step A7 shown in FIG. 10 is executed, the process proceeds to step D1. Steps D1 to D5 and D10 are the same as steps B1 to B5 and B7 shown in FIG.

  In step D6, the replacement determination unit 22b determines whether similarity equal to or greater than a predetermined threshold is obtained from the comparison of feature amounts. When the similarity calculated in step D5 is equal to or greater than a predetermined threshold, it is determined from the feature amount comparison that similarity equal to or greater than the predetermined threshold is obtained, and the process proceeds to step D3. When the degree of similarity calculated in step D5 is less than a predetermined threshold, it is determined from the feature amount comparison that no similarity equal to or higher than the predetermined threshold has been obtained, and the process proceeds to step D7.

  In step D7, the replacement determination unit 22b determines whether or not the number of characters of the target word is equal to or greater than a predetermined reference number M. When the number of characters of the target word is greater than or equal to the predetermined reference character number M, the process proceeds to step D8, and when the number of characters of the target word is less than the predetermined reference character number M, the process proceeds to step D10. The replacement determination unit 22b determines not to replace when proceeding to step D10. In step D8, the replacement determination unit 22b resets the predetermined threshold value to a low value.

  In step D9, the replacement determination unit 22b determines whether or not the similarity is equal to or higher than a preset threshold value. When the similarity is equal to or higher than the preset predetermined threshold, the process proceeds to step D3, and when the similarity is less than the preset predetermined threshold, the process proceeds to step D10. The replacement determination unit 22b determines to replace when proceeding to step D3, and determines not to replace when proceeding to step D10.

  Step D1 shown in FIG. 25 corresponds to a word string generation process, steps D2 and D4 to D9 shown in FIG. 25 correspond to a replacement determination process, and steps D3 and D10 shown in FIG. It corresponds to.

  Thus, in the first to third embodiments described above, a recognized character string can be obtained by a simple process without using a dictionary other than the character recognition dictionary 14 and the word dictionary 24, for example, another dictionary such as a similarity character dictionary. The error in can be corrected to the correct character. Since no dictionary other than the character recognition dictionary 14 and the word dictionary 24 is required, an increase in storage capacity can be avoided.

  In the first to third embodiments described above, when the character in the recognition result character string is different from the character in the word character string, the corresponding character number in the recognition result character string is changed for the different character. Depending on the similarity calculated from the feature data extracted from the character image and the feature data in the character recognition dictionary corresponding to the character in the word character string, the character in the recognition result character string is replaced with the character in the word character string. Although it was set as the structure which determines whether it replaces, it is not limited to this. For example, the character feature data recorded in the character recognition dictionary 14 corresponding to the first candidate candidate character in the recognition result character string, and the character recorded in the character recognition dictionary 14 corresponding to the character in the word character string. Depending on the similarity calculated from the feature data, a character in the recognition result character string may be replaced with a character in the word character string.

  In the first to third embodiments described above, simple similarity is used as the similarity obtained by comparing the character feature amounts and quantifying how much the characters are similar to each other. However, the present invention is not limited to this. For example, it is also possible to use composite similarity calculated by further calculating from simple similarity.

  In the first to third embodiments described above, an example in which a document in which a document is written is read, a character string of the document represented in the image of the read document is divided into words, and collect processing is performed is shown. However, the present invention is not limited to this. For example, it is also applicable when reading a character string describing an address and replacing the word dictionary with an address dictionary to perform a collect process. It is possible to correct a misrecognized address to a correct address, that is, to correct it. . Even when the word dictionary is replaced with the address dictionary and applied to the address character string to perform the collect process, no other dictionary such as a similarity character dictionary is required, so that an additional capacity of the memory 13 is not required. In addition, correct collection processing can be performed by a simple method.

  In the above-described embodiment, the input / output control unit 12 is controlled, and the character string row cutout unit 15, the character cutout unit 16, the feature extraction unit 17, the matching unit 18, the character string generation unit 19, and the post-processing units 20 and 20a. , 20b and the like are stored in a storage device such as the memory 13 of the information processing apparatus 1, 1a, 1b, but are not limited to this storage device, and are a computer-readable recording medium May be recorded. The recording medium may be, for example, a recording medium that is readable by providing a program reading device as an external storage device (not shown) in the information processing apparatus 1, 1a, 1b and inserting the recording medium therein. It may be a storage device of the device.

  Any recording medium may be used as long as the stored program is accessed from a computer and executed. Alternatively, any recording medium may be configured such that the program is read, the read program is stored in the program storage area of the storage device, and the program is executed.

The recording medium configured to be separable from the information processing apparatuses 1, 1a, 1b is, for example, a tape-type recording medium such as a magnetic tape / cassette tape, a magnetic disk such as a flexible disk / hard disk, or a CD-ROM (Compact Disk Read Only). Memory) / MO (
Magnet-type optical disks (MDs) / MDs (Mini Discs) / DVDs (Digital Versatile Disks) / CD-Rs (Compact Disk Recordables) / Blu-rays and other optical discs, IC (Integrated Circuit) cards (including memory cards) ) / Card-type recording media such as optical cards, or mask ROM / EPROM (Erasable Programmable)
Read Only Memory) / EEPROM (Electrically Erasable Programmable Read Only
Memory) / a recording medium that carries a fixed program including a semiconductor memory such as a flash ROM.

The information processing apparatuses 1, 1a, and 1b may be configured to be connectable to a communication network, and the program may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN (Local Area Network), ISDN (Integrated Services Digital Network), VAN (Value Added Network), CATV (Community Antenna Television) communication network. A communication network such as a virtual private network, a telephone line network, a mobile communication network, or a satellite communication network can be used. Also, the transmission medium constituting the communication network is not particularly limited. For example, IEEE1394, USB (Universal Serial Bus), power line carrier, cable TV line, telephone line, ADSL (Asymmetric Digital Subscriber)
Line (wire) lines, etc., IrDA (Infrared Data Association) or infrared light used for remote control, Bluetooth (registered trademark), 802.11 wireless, HDR (High Data Rate), mobile phone network, satellite line, terrestrial digital It can also be used wirelessly such as on the network. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program is embodied by electronic transmission.

  As described above, the memory 13 stores the character recognition dictionary 14 that records the character and the feature value obtained by quantifying the feature of the character in association with each other, and the word dictionary 24 that records the word composed of at least one character. The character string line cutout unit 15 cuts out a character string image representing a character string for each line from an image of a document on which a character string line that is a character string is described. The character cutout unit 16 cuts out character images representing each character one by one from the character string image cut out by the character string row cutout unit 15. The feature extraction unit 17 extracts the feature amount of the character represented by each character image from the character image cut out by the character cutout unit 16. The matching unit 18 compares the feature amount extracted by the feature extraction unit 17 with the feature amount recorded in the character recognition dictionary 14 stored in the memory 13, and for each character represented by the character image, the character represented by the character image. At least one candidate character that can be recognized as is selected from the character recognition dictionary 14. The character string generation unit 19 generates a recognition result character string by combining candidate characters for each character image selected by the matching unit 18. The word string generation unit 21 collates the candidate characters constituting the recognition result character string generated by the character string generation unit 19 with the words recorded in the word dictionary 24 stored in the memory 13, and stores them in the word dictionary 24. A character string in which all the characters match the recorded word or a word in which only a part of the characters match is extracted, and a word character string including the extracted words is generated. The replacement determination units 22, 22 a, and 22 b differ by comparing the recognition result character string generated by the character string generation unit 19 and the word character string generated by the word string generation unit 21 for each character at a corresponding position. When there is a character, the different character matches the character of the word character string among the feature amount extracted by the feature extraction unit 17 and the character feature amount recorded in the character recognition dictionary 14 stored in the memory 13. It compares with the feature-value memorize | stored matched with a character, and when a comparison result satisfy | fills the comparison criterion defined beforehand, it determines with replacing the character in a recognition result character string with the corresponding character in a word character string. Then, the replacement unit 23 replaces the character in the recognition result character string determined to be replaced by the replacement determination unit 22, 22a, 22b with the character at the corresponding position in the word character string.

  Therefore, it is possible to correct a recognized character error to a correct character without using a dictionary other than the character recognition dictionary 14 and the word dictionary 24. Since no dictionary other than the character recognition dictionary 14 and the word dictionary 24 is required, an increase in storage capacity can be avoided.

  Furthermore, the memory 13 stores a character recognition dictionary 14 that records characters and feature values obtained by quantifying the features of the characters in association with each other, and a word dictionary 24 that records words composed of at least one character. The character string line cutout unit 15 cuts out a character string image representing a character string for each line from an image of a document on which a character string line that is a character string is described. The character cutout unit 16 cuts out character images representing each character one by one from the character string image cut out by the character string row cutout unit 15. The feature extraction unit 17 extracts the feature amount of the character represented by each character image from the character image cut out by the character cutout unit 16. The matching unit 18 compares the feature amount extracted by the feature extraction unit 17 with the feature amount recorded in the character recognition dictionary 14 stored in the memory 13, and for each character represented by the character image, the character represented by the character image. At least one candidate character that can be recognized as is selected from the character recognition dictionary 14. The character string generation unit 19 generates a recognition result character string by combining candidate characters for each character image selected by the matching unit 18. The word string generation unit 21 collates the candidate characters constituting the recognition result character string generated by the character string generation unit 19 with the words recorded in the word dictionary 24 stored in the memory 13, and stores them in the word dictionary 24. A character string in which all the characters match the recorded word or a word in which only a part of the characters match is extracted, and a word character string including the extracted words is generated. The replacement determination units 22, 22 a, and 22 b differ by comparing the recognition result character string generated by the character string generation unit 19 and the word character string generated by the word string generation unit 21 for each character at a corresponding position. When there is a character, for different characters, the feature amount stored in association with the character that matches the character of the recognition result character string among the feature amounts recorded in the character recognition dictionary 14 stored in the memory 13, and the memory 13 is compared with the feature amount stored in association with the character that matches the character of the word character string among the feature amounts of the characters recorded in the character recognition dictionary 14 stored in 13, and a comparison criterion determined in advance by the comparison result is determined. When it is satisfied, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string. Then, the replacement unit 23 replaces the character in the recognition result character string determined to be replaced by the replacement determination unit 22, 22a, 22b with the character at the corresponding position in the word character string.

  Therefore, it is possible to correct a recognized character error to a correct character without using a dictionary other than the character recognition dictionary 14 and the word dictionary 24. Since no dictionary other than the character recognition dictionary 14 and the word dictionary 24 is required, an increase in storage capacity can be avoided.

  Furthermore, the replacement determination units 22, 22 a, and 22 b compare the recognition result character string generated by the character string generation unit 19 and the word character string generated by the word string generation unit 21 for each character at a corresponding position. If there is a different character, the different character matches the character of the word character string among the feature amount extracted by the feature extraction unit 17 and the feature amount recorded in the character recognition dictionary 14 stored in the memory 13. The similarity is calculated by comparing the feature quantity stored in association with the character, and when the calculated similarity is equal to or greater than a predetermined threshold, the character in the recognition result character string is converted to the corresponding character in the word character string. It is determined to be replaced with. Therefore, since the similarity based on the feature amount is used as an index, the similarity of characters can be determined with a simple calculation.

  Furthermore, the replacement determination units 22, 22 a, and 22 b compare the recognition result character string generated by the character string generation unit 19 and the word character string generated by the word string generation unit 21 for each character at a corresponding position. If there is a different character, the different character is stored in association with the character that matches the character of the recognition result character string among the feature values recorded in the character recognition dictionary 14 stored in the memory 13. The similarity is calculated by comparing the feature quantity recorded in the character recognition dictionary 14 stored in the memory 13 with the feature quantity stored in association with the matching character of the word character string. When the similarity is equal to or greater than a predetermined threshold, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string. Therefore, since the similarity based on the feature amount is used as an index, the similarity of characters can be determined with a simple calculation.

  Further, the replacement determination unit 22, 22a, 22b is configured to store the words recorded in the word dictionary 24 when the word string generation unit 21 generates the word character string when the calculated similarity is less than a predetermined threshold. And a character string that matches all characters, or a word that matches only a part of the characters and is not included in the word character string, the calculated similarity is a predetermined threshold value. When the value is equal to or greater than a predetermined second threshold value that is less than the value, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string.

  Therefore, when only one word candidate is generated, the word candidate is likely to be correct and can be replaced as an optimum word even if the similarity is less than a predetermined threshold. . For example, in optical character recognition, if one character of a word consisting of a plurality of characters is not correctly recognized due to “faint” or “smear”, the degree of similarity will be low and will be less than a predetermined threshold. However, if there is only one word extracted from the word dictionary 24, there is a very high possibility that it is a correct answer, and there is an effect that it can be corrected to a correct character by correcting using this word. is there.

  Furthermore, when the calculated similarity is less than a predetermined threshold and the number of characters of the word including the different characters is greater than or equal to a predetermined reference number, the replacement determination unit 22, 22a, 22b Is greater than or equal to a predetermined second threshold value that is a value less than a predetermined threshold value, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string. Therefore, for words with a large number of characters, even if the degree of similarity is less than a predetermined threshold value, there is a high possibility that the word is correct, and an optimal word can be selected and replaced.

  Furthermore, the information processing apparatus includes a memory 13 for storing a character recognition dictionary 14 that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a word dictionary 24 that records a word composed of at least one character. Thus, when recognizing characters using the character recognition dictionary 14 and the word dictionary 24 stored in the memory 13, in step A2 shown in FIG. 10, characters are extracted from an image of a document in which character string rows that are character strings are described. A character string image representing a column is cut out for each row. In step A3 shown in FIG. 10, a character image representing each character is cut out character by character from the character string image cut out in step A2 shown in FIG. In step A4 shown in FIG. 10, the feature amount of the character represented by each character image is extracted from the character image cut out in step A3 shown in FIG. In step A5 shown in FIG. 10, the feature quantity extracted in step A4 shown in FIG. 10 is compared with the feature quantity recorded in the character recognition dictionary 14 stored in the memory 13, and each character represented by the character image is displayed. Then, at least one candidate character that can be recognized as the character represented by the character image is selected from the character recognition dictionary 14. In step A6 shown in FIG. 10, a recognition result character string is generated by combining candidate characters for each character image selected in step A5 shown in FIG. In step B1 shown in FIG. 11, step C1 and C2 shown in FIG. 17, or step D1 shown in FIG. 25, candidate characters constituting the recognition result character string generated in step A6 shown in FIG. The words recorded in the word dictionary 24 stored in the memory 13 are collated, and a character string in which all characters match the word recorded in the word dictionary 24 or a word in which only some characters match is extracted, Generate a word string consisting of the extracted words. In steps B2, B4 to B6 shown in FIG. 11, steps C3, C5 to C8 shown in FIG. 17, or steps D2, D4 to D9 shown in FIG. 25, the recognition generated in step A6 shown in FIG. The result character string and the word character string generated in step B1 shown in FIG. 11, step C1, C2 shown in FIG. 17, or step D1 shown in FIG. 25 are compared for each character at the corresponding position. In the case where there are different characters, for the different characters, among the feature values extracted in step A4 shown in FIG. 10 and the character feature values recorded in the character recognition dictionary 14 stored in the memory 13, The feature quantity stored in association with the character that matches the character is compared, and when the comparison result satisfies a predetermined comparison criterion, the character in the recognition result character string is replaced with the corresponding character in the word character string. Replace the judges. Then, in steps B3 and B7 shown in FIG. 11, steps C4 and C9 shown in FIG. 17, or steps D3 and D10 shown in FIG. 25, steps B2, B4 to B6 shown in FIG. The characters in the recognition result character string determined to be replaced in steps C3, C5 to C8 or steps D2, D4 to D9 shown in FIG. 25 are replaced with the characters at the corresponding positions in the word character string.

  Therefore, it is possible to correct a recognized character error to a correct character without using a dictionary other than the character recognition dictionary 14 and the word dictionary 24. Since no dictionary other than the character recognition dictionary 14 and the word dictionary 24 is required, an increase in the storage capacity of the information processing apparatus can be avoided.

  Furthermore, the information processing apparatus includes a memory 13 for storing a character recognition dictionary 14 that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a word dictionary 24 that records a word composed of at least one character. Thus, when recognizing characters using the character recognition dictionary 14 and the word dictionary 24 stored in the memory 13, in step A2 shown in FIG. 10, characters are extracted from an image of a document in which character string rows that are character strings are described. A character string image representing a column is cut out for each row. In step A3 shown in FIG. 10, a character image representing each character is cut out character by character from the character string image cut out in step A2 shown in FIG. In step A4 shown in FIG. 10, the feature amount of the character represented by each character image is extracted from the character image cut out in step A3 shown in FIG. In step A5 shown in FIG. 10, the feature quantity extracted in step A4 shown in FIG. 10 is compared with the feature quantity recorded in the character recognition dictionary 14 stored in the memory 13, and each character represented by the character image is displayed. Then, at least one candidate character that can be recognized as the character represented by the character image is selected from the character recognition dictionary 14. In step A6 shown in FIG. 10, a recognition result character string is generated by combining candidate characters for each character image selected in step A5 shown in FIG. In step B1 shown in FIG. 11, step C1 and C2 shown in FIG. 17, or step D1 shown in FIG. 25, candidate characters constituting the recognition result character string generated in step A6 shown in FIG. The words recorded in the word dictionary 24 stored in the memory 13 are collated, and a character string in which all characters match the word recorded in the word dictionary 24 or a word in which only some characters match is extracted, Generate a word string consisting of the extracted words. In steps B2, B4 to B6 shown in FIG. 11, steps C3, C5 to C8 shown in FIG. 17, or steps D2, D4 to D9 shown in FIG. 25, the recognition generated in step A6 shown in FIG. The result character string and the word character string generated in step B1 shown in FIG. 11, step C1, C2 shown in FIG. 17, or step D1 shown in FIG. 25 are compared for each character at the corresponding position. If there are different characters, for the different characters, among the feature amounts recorded in the character recognition dictionary 14 stored in the memory 13, a feature amount stored in association with a character that matches the character of the recognition result character string; Of the character feature quantities recorded in the character recognition dictionary 14 stored in the memory 13, the feature quantities stored in association with the characters matching the characters in the word character string are compared, and the comparison result is determined in advance. When meeting the comparison reference that determines that replaces the character recognition result in the character string in the corresponding character in the word string. Then, in steps B3 and B7 shown in FIG. 11, steps C4 and C9 shown in FIG. 17, or steps D3 and D10 shown in FIG. 25, steps B2, B4 to B6 shown in FIG. The characters in the recognition result character string determined to be replaced in steps C3, C5 to C8 or steps D2, D4 to D9 shown in FIG. 25 are replaced with the characters at the corresponding positions in the word character string.

  Therefore, it is possible to correct a recognized character error to a correct character without using a dictionary other than the character recognition dictionary 14 and the word dictionary 24. Since no dictionary other than the character recognition dictionary 14 and the word dictionary 24 is required, an increase in the storage capacity of the information processing apparatus can be avoided.

  Furthermore, it can provide as a program for making a computer perform each process of the information processing method.

  Furthermore, the present invention can be provided as a computer-readable recording medium that records a program for causing a computer to execute each step of the information processing method.

1, 1a, 1b Information processing device 2 Input device 3 Output device 11 Control unit 12 Input / output control unit 13 Memory 14 Character recognition dictionary 15 Character string row cutout unit 16 Character cutout unit 17 Feature extraction unit 18 Matching unit 19 Character string generation unit 20, 20a, 20b Post-processing unit 21 Word string generation unit 22, 22a, 22b Replacement determination unit 23 Replacement unit 24 Word dictionary 30 Character 31 Mesh 32 Feature amount 33 Mesh feature amount

Claims (11)

A storage unit for storing a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other;
A character string row cutout unit that cuts out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout unit that cuts out one character image representing each character from the character string image cut out by the character string line cutout unit;
A feature extraction unit that extracts a feature amount of a character represented by each character image from the character image cut out by the character cutout unit;
The feature amount extracted by the feature extraction unit is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection unit for selecting at least one from the character recognition dictionary;
A character string generation unit that generates a recognition result character string by combining candidate characters for each character image selected by the selection unit;
The candidate characters constituting the recognition result character string generated by the character string generation unit are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation unit that extracts a character string or a word that matches only some characters and generates a word character string composed of the extracted words;
If there is a different character by comparing the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at the corresponding position, feature extraction is performed for the different characters The feature amount extracted by the unit is compared with the feature amount stored in association with the character that matches the character of the word character string among the character feature amounts recorded in the character recognition dictionary stored in the storage unit. A replacement determination unit that determines whether to replace a character in the recognition result character string with a corresponding character in the word character string based on the comparison result;
An information processing apparatus comprising: a replacement unit that replaces a character in a recognition result character string determined to be replaced by a replacement determination unit with a character at a corresponding position in a word character string. A storage unit for storing a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other;
A character string row cutout unit that cuts out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout unit that cuts out one character image representing each character from the character string image cut out by the character string line cutout unit;
A feature extraction unit that extracts a feature amount of a character represented by each character image from the character image cut out by the character cutout unit;
The feature amount extracted by the feature extraction unit is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection unit for selecting at least one from the character recognition dictionary;
A character string generation unit that generates a recognition result character string by combining candidate characters for each character image selected by the selection unit;
The candidate characters constituting the recognition result character string generated by the character string generation unit are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation unit that extracts a character string or a word that matches only some characters and generates a word character string composed of the extracted words;
When there is a different character by comparing the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at the corresponding position, the storage unit Among the feature quantities recorded in the character recognition dictionary stored in the memory, and the feature quantities stored in association with the characters matching the characters in the recognition result character string, and the features recorded in the character recognition dictionary stored in the storage unit The feature amount stored in association with the character that matches the character of the word character string is compared, and when the comparison result satisfies a predetermined comparison criterion, the character in the recognition result character string is A replacement determination unit that determines to replace with a corresponding character;
An information processing apparatus comprising: a replacement unit that replaces a character in a recognition result character string determined to be replaced by a replacement determination unit with a corresponding character in a word character string.   When the replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position and has different characters The different characters are stored in association with the character extracted from the feature extraction unit and the character matching the character of the word character string among the feature values recorded in the character recognition dictionary stored in the storage unit. The similarity is calculated by comparing with the feature amount to be determined, and when the calculated similarity is equal to or greater than a predetermined threshold, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string The information processing apparatus according to claim 1.   When the replacement determination unit compares the recognition result character string generated by the character string generation unit and the word character string generated by the word string generation unit for each character at a corresponding position and has different characters For different characters, feature quantities stored in association with characters that match the characters in the recognition result character string among the feature quantities recorded in the character recognition dictionary stored in the storage unit, and stored in the storage unit Among the feature quantities recorded in the character recognition dictionary to be compared with the feature quantity stored in association with the character that matches the character of the word character string, and the similarity is calculated, and the calculated similarity is a predetermined threshold value When it is above, it determines with replacing the character in a recognition result character string with the corresponding character in a word character string, The information processing apparatus of Claim 2 characterized by the above-mentioned.   When the calculated similarity is less than a predetermined threshold, when the word string generation unit generates the word character string, the replacement determination unit matches all words with the words recorded in the word dictionary. If a word that matches only a character string or a part of characters and a word that is not included in the word character string has been extracted, the calculated similarity is a value that is less than a predetermined threshold value. 5. The information processing apparatus according to claim 3, wherein when it is equal to or more than a predetermined second threshold value, it is determined that the character in the recognition result character string is replaced with the corresponding character in the word character string.   When the calculated similarity is less than a predetermined threshold, and the number of characters of the word including the different characters is equal to or greater than a predetermined reference number, the calculated similarity is less than a predetermined threshold 5. The information according to claim 3, wherein when the value is equal to or greater than a predetermined second threshold value, the character in the recognition result character string is determined to be replaced with the corresponding character in the word character string. Processing equipment. An information processing apparatus including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character. An information processing method for recognizing characters using a character recognition dictionary and a word dictionary stored in
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
If there are different characters by comparing the recognition result character string generated in the character string generation process and the word character string generated in the word string generation process for each character at the corresponding position, feature extraction is performed for the different characters The feature amount extracted in the process is compared with the feature amount stored in association with the character matching the character of the word character string among the character feature amounts recorded in the character recognition dictionary stored in the storage unit. A replacement determination step for determining that a character in the recognition result character string is replaced with a corresponding character in the word character string when the comparison result satisfies a predetermined comparison criterion;
A replacement step of replacing a character in the recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in the word character string. An information processing apparatus including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word composed of at least one character. An information processing method for recognizing characters using a character recognition dictionary and a word dictionary stored in
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
When there are different characters by comparing the recognition result character string generated in the character string generation step and the word character string generated in the word string generation step for each character at the corresponding position, the storage unit Among the feature quantities recorded in the character recognition dictionary stored in the memory, and the feature quantities stored in association with the characters matching the characters in the recognition result character string, and the characters recorded in the character recognition dictionary stored in the storage unit When the comparison result meets a predetermined comparison criterion, the character in the recognition result character string is converted to the word character string. A replacement determination step for determining to replace with the corresponding character in
A replacement step of replacing a character in the recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in the word character string. In a computer including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word including at least one character,
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
If there are different characters by comparing the recognition result character string generated in the character string generation process and the word character string generated in the word string generation process for each character at the corresponding position, feature extraction is performed for the different characters The feature amount extracted in the process is compared with the feature amount stored in association with the character matching the character of the word character string among the character feature amounts recorded in the character recognition dictionary stored in the storage unit. A replacement determination step for determining that the character in the recognition result character string is replaced with the corresponding character in the word character string when the comparison result satisfies a predetermined comparison criterion;
A program for executing a replacement step of replacing a character in a recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in a word character string. In a computer including a character recognition dictionary that records a character and a feature value obtained by quantifying the character of the character in association with each other, and a storage unit that stores a word dictionary that records a word including at least one character,
A character string row cutting step for cutting out a character string image representing a character string for each line from an image of a document in which character string lines that are character strings are described;
A character cutout step of cutting out one character image representing each character from the character string image cut out in the character string row cutout step;
A feature extraction step of extracting the feature amount of the character represented by each character image from the character image cut out in the character cutout step;
The feature amount extracted in the feature extraction step is compared with the feature amount recorded in the character recognition dictionary stored in the storage unit, and for each character represented by the character image, a candidate character that can be recognized as the character represented by the character image is determined. A selection step of selecting at least one from the character recognition dictionary;
A character string generation step of generating a recognition result character string by combining candidate characters for each character image selected in the selection step;
The candidate characters constituting the recognition result character string generated in the character string generation step are compared with the words recorded in the word dictionary stored in the storage unit, and the words recorded in the word dictionary match all the characters. A word string generation step of extracting a word string or a word that matches only some characters, and generating a word string composed of the extracted words;
When there are different characters by comparing the recognition result character string generated in the character string generation step and the word character string generated in the word string generation step for each character at the corresponding position, the storage unit Among the feature quantities recorded in the character recognition dictionary stored in the memory, and the feature quantities stored in association with the characters matching the characters in the recognition result character string, and the characters recorded in the character recognition dictionary stored in the storage unit When the comparison result meets a predetermined comparison criterion, the character in the recognition result character string is converted to the word character string. A replacement determination step of determining whether to replace with a corresponding character in
A program for executing a replacement step of replacing a character in a recognition result character string determined to be replaced in the replacement determination step with a character at a corresponding position in a word character string.   The computer-readable recording medium which recorded the program of Claim 9 or 10.

Images