JPH09231318A - Character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide dictionary retrieval capable of absorbing the variation of uppercase letter/lowercase letter without increasing the size of a collating dictionary. SOLUTION: The character recognizing device recognizes a character string from picture information, collates a candidate lattice storing each recognized character and its candidate characters with a word dictionary by a language processing part 106 and outputs a finally determined character string. The language processing part 106 is provided with an English candidate lattice preparing part 108 for preparing an English candidate lattice capable of storing candidate characters from the 1st candidate up to the lowest candidate by maximum word length in order to collate an English word with a dictionary and an alphabetical uppercase letter/lowercase letter adding part 109 for adding an uppercase letter or a lowercase letter to an English candidate lattice so as to provide both the uppercase and lowercase letters of the character in a pair.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for a document in which Japanese and English are mixed, and more specifically, it performs language processing on a candidate lattice including each candidate character for a recognized character string. The present invention relates to a character recognition device that outputs a character string finally determined by the language processing.

[0002]

2. Description of the Related Art A conventional general character recognition device is
The language processing unit is provided for post-processing post-recognition to make the recognized character string a more plausible sentence. The language processing unit searches the candidate lattice of the recognition result by using a word dictionary and determines the candidate word, and determines it. Then, as a result of this processing, if there is a defect in the first candidate character of the recognition result, processing such as replacing with the character after the second candidate character is executed.

Further, a technique related to a dictionary search device in the field of machine translation or the like is disclosed in Japanese Patent Laid-Open Nos. 61-70662 and 62-82466. Here, a means for converting uppercase letters into lowercase letters in the input character string is provided, and when there is no corresponding word in a normal word search, the word dictionary is searched by the character string with the uppercase letters converted to lowercase letters.

[0004]

However, in the conventional apparatus as described above, the following English character strings have a problem in collating with the English word dictionary.

First, for example, "window", "W"
There are various notation methods even for the same word, such as only uppercase letters, lowercase letters, or a mixture of uppercase letters and lowercase letters, such as "INDOW" and "Windows".

Secondly, not all words allow plural notations, and words that are described only in capital letters, such as abbreviations in which "INS" is not described as "ins", etc. is there.

Thirdly, proper names such as place names and personal names are generally not written in lower case only. For example, the place name "E
"ngland" is not described as "england", and the personal name "John" is not described as "john".

Therefore, a method of creating a word dictionary that describes all the variations of the notations as shown in these three and collating with a candidate lattice can be considered. However, this method requires a large dictionary memory capacity for collation. Therefore, there is a problem that the memory cost increases and the dictionary search speed decreases.

The present invention has been made in view of the above, and it is an object of the present invention to realize a dictionary search that absorbs fluctuations in uppercase and lowercase letters without increasing the dictionary size for collation.

[0010]

In order to achieve the above object, in the character recognition device according to claim 1, a character string is recognized from image information, and the recognition result character and its candidate are recognized by the language processing means. In a character recognition device for collating a candidate lattice containing characters with a word dictionary and outputting a finally determined character string, the language processing means includes first to lower ranks for collating English words in a dictionary. If a candidate character has an alphabetic character, the candidate lattice creating means for creating an alphabetic character candidate capable of storing the candidate character in the maximum word length, and if the candidate character has an alphabetic character, both the uppercase and lowercase letters of the character are paired. It is provided with means for adding uppercase / lowercase letters in the candidate lattice.

That is, when the candidate lattice is matched with the dictionary in the language processing means, if there is an alphabetic character in the candidate character, uppercase letters or lowercase letters are included in the candidate lattice so that both uppercase letters and lowercase letters of the letters become a pair. By adding "," both uppercase and lowercase letters are always candidate characters, so it is possible to perform word matching even if the written description in the actual manuscript is not in the English word dictionary.

Therefore, it is not necessary to take measures such as increasing the number of entries in the English word dictionary depending on the notation,
There is no need to increase the dictionary size unnecessarily. In addition, since the dictionary search for English words can be performed without being aware of uppercase and lowercase letters, it is not necessary to newly prepare a dictionary matching algorithm dedicated to English words.

Further, in the character recognition device according to the second aspect, the language processing means recognizes each notation of the selected word as a result of the matching between the candidate lattice to which the candidate character is added and the word dictionary. It further comprises English uppercase / lowercase correction means for correcting uppercase / lowercase based on the order of candidate characters.

That is, as a result of collation between the candidate lattice to which the candidate character is added and the word dictionary, the uppercase and lowercase letters of the English letters constituting the collated English word notation are determined based on the candidate rank of the recognized character. Because of the replacement, the word notation described in the actual manuscript is approximated, and the recognition accuracy is improved.

Further, in the character recognition device according to the third aspect, the language processing means, as a result of collating the candidate lattice to which the candidate character has been added with the word dictionary, all the notation of the selected word is capitalized. In this case, an English word determination means for determining whether or not the word should be evaluated based on the candidate rank of each recognized character is further provided.

That is, as a result of collating the candidate lattice to which the candidate character is added with the word dictionary, if the selected word is in all capital letters, it should be evaluated based on the candidate rank of each recognized character. By determining whether or not it is a word, unnecessary words such as abbreviations can be deleted from the evaluation target, which improves processing accuracy and reduces linguistic analysis errors. Accuracy is improved.

Further, in the character recognition device according to the fourth aspect, the language processing means, as a result of collating the candidate lattice to which the candidate character is added with the word dictionary, the selected word is capitalized. In the case of mixed words in lower case, the apparatus further comprises English word determination means for determining whether or not the word should be evaluated based on the candidate rank of each recognized character.

That is, as a result of matching the candidate lattice to which the candidate character is added with the word dictionary, if the selected word is a mixed word of uppercase and lowercase letters, it is evaluated based on the candidate rank of each recognized character. By determining whether or not a word should be a target, unnecessary words can be deleted from the evaluation target, such as a place name and a person's name, so that processing accuracy is improved and language analysis is performed. Errors are reduced and recognition accuracy is improved.

In addition, in the character recognition device according to the fifth aspect, the language processing means does not describe all upper and lower case fluctuations of the word notation, and the language processing means does not include one for each word.
It further comprises a word collating means for collating the candidate lattice and the word dictionary with an English word dictionary described in one notation to obtain a candidate word.

That is, by having an English word dictionary in which each word is described in one notation without describing all upper and lower case fluctuations in the word notation, the size of the word dictionary is not increased, and the dictionary entries are This saves the effort of creating multiple entries and reduces the work of creating entries.

In the character recognition device according to the sixth aspect, the language processing means has a common word dictionary matching algorithm for both Japanese and English character strings.

That is, since the word dictionary matching algorithm is shared in both Japanese and English character strings and it is not necessary to prepare a new dictionary matching algorithm dedicated to English words, a mixed document of Japanese and English is handled. In OCR, etc., simplification and speeding up of processing are realized.

[0023]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the accompanying drawings.

[Embodiment] (Configuration of Embodiment) FIG. 1 is a block diagram showing an overall configuration of a character recognition apparatus according to an embodiment. In the figure, the character recognition device includes an image input unit 101 for inputting an image image to be recognized, a line cutout unit 102 for extracting an image image in units of lines, and character information of line parts extracted by the line cutout unit 102. Character extraction unit 1 to extract
03, a character recognition processing unit 104 that recognizes the character information cut out by the character cutting unit 103, and a character recognition processing unit 1
A character combination selection processing unit 105 for combining and selecting the characters recognized in 04, and a language processing unit (language processing unit for performing language processing on a candidate lattice including each candidate character after recognition processing as described later ( Post-processing unit) 106
And a fixed character string output unit 10 for fixing the character information processed by the language processing unit (post-processing unit) 106 and outputting it as a character string.
7 is comprised.

The language processing unit (post-processing unit) 106 is
In addition to the normal candidate lattice, an English character candidate lattice creating unit 108 as an English character candidate lattice creating means for creating an English character candidate lattice for dictionary matching of English words, and when an English character exists in the candidate character, The candidate lattice and the word dictionary are collated with an uppercase / lowercase addition unit 109 as an uppercase / lowercase addition means for adding uppercase or lowercase letters in the candidate lattice so that both uppercase and lowercase letters of the character are paired, A word matching unit 110 as a word matching means for obtaining a candidate word, and for a word whose dictionary matching is not in lowercase, determine the accuracy of the character based on the difference in the candidate rank of the recognized character to determine whether it is a candidate word. An English word determination unit 111 as an English word determination unit for determining and an upper case / lower case correction of the notation of the selected word are performed based on the candidate rank of each recognized character. It is composed of an English upper / lower case correction unit 112 as an upper / lower case correction unit, and a candidate word evaluation unit 113 which determines whether or not the word should be evaluated based on the candidate rank of each recognized character. There is.

(Operation of Embodiment) Next, the operation of the character recognition device configured as described above will be described. First, the image input unit 101 inputs an image image, character recognition of this image image is performed, and a candidate lattice containing the recognized character and its candidate character is created. An example of this normal candidate lattice is shown in FIG.

The operation of the language processing unit (post-processing unit) 106 will be described below with reference to the flowchart shown in FIG. In the figure, first, it is determined whether or not there is an alphabetic character at the character position at the beginning of the target (S301). Here, not only the first candidate character but also the lower candidate characters are checked. If it is determined in this step S301 that there is an alphabetic character at the beginning character position, the alphabetic character candidate lattice creating unit 108 executes the processing operation (S302), and further the English uppercase / lowercase character adding unit 109 executes the processing operation. To do.

On the other hand, when it is determined in step S301 that there is no alphabetic character at the leading character position, or after the processing operation by the uppercase / lowercase addition unit 109 in step S303 is performed, the normal word matching unit 110 is used. The process is executed (S304). The operations of the English character candidate lattice creation unit 108, the English uppercase / lowercase addition unit 109, and the word matching unit 110 will be described in detail below.

In addition to the normal candidate lattice, the English character candidate lattice creating unit 108 can store the candidate characters from the first candidate to the lower candidate for the maximum word length in order to match the dictionary of English words. To create. Also, if there is an alphabetic character in the candidate character, a set of candidate characters is copied to the English character candidate lattice, and the process ends when there are no more alphabetic characters in the candidate. This example is shown in FIG. After that, the processing shifts to the processing operation by the English uppercase / lowercase addition unit 109.

The English upper / lower case addition unit 109 checks whether or not both upper and lower case letters of the alphabet exist in the candidate characters from the first candidate to the lower order with respect to the alphabet candidate lattice. If it does not exist, the missing character is added to the last position of the candidate rank as in the example shown in FIG.
Here, upper and lower case letters are forcibly added regardless of the character shape, such as n and N and D and d.

The word collating unit 110 collates the candidate lattice with the word dictionary to obtain candidate words. If the English character candidate lattice exists, the English character candidate lattice is compared with the English word dictionary. On the other hand, if the English character candidate lattice does not exist, the normal candidate lattice and the Japanese dictionary are matched.

As the matching algorithm, a generally known method such as the A * algorithm may be used. Also, the word dictionary may use two sets of Japanese and English, or merge two (1).
It may be a dictionary.

FIG. 6 shows an example of entries in this English word dictionary. General words are in lower case, abbreviations are in upper case,
Only the first letter of a place name or person's name is capitalized. In this way, one word has only one notation as an entry. And
If the searched word is an English word, go to the English word determination unit 111,
Otherwise, the candidate word is set as a candidate word, and the candidate word evaluation unit 113 performs evaluation such as cost calculation.

That is, word matching processing is performed in step 304, and it is determined whether the candidate word is an English word (S305). If it is determined that the candidate word is an English word, it is further determined whether the notation is lowercase (S306). Then, when it is determined that the notation is not lowercase, the process proceeds to the processing operation (S307) by the English word determination unit 111.

As shown in FIG. 7, the English word determination unit 111 determines the accuracy of a character based on the difference in the candidate character candidate rank for words whose dictionary collation is not lowercase, and regards it as a candidate word. Determine if it should. For a word in all alphabetic characters, such as "INS", if there is at least one character whose lowercase letter is lower in candidate order than uppercase letter, the word is invalid and set. do not do.

For a word in which the first character is uppercase and the second and subsequent letters are lowercase, for example, "Iran", etc., only the first character is evaluated, and lowercase letters come first before uppercase letters. If so, the word is invalid and is not set. That is, it is determined whether the determination result by the English word determination unit 111 is OK (S308). If it is determined to be OK, words that do not correspond to these are input to the upper / lower case correction unit 112, and the processing operation by the upper / lower case correction unit 112 is performed (S309). If it is determined in step 308 that the word is not OK, the word is deleted (S310).

As shown in FIG. 8, the upper / lower case correction unit 112 searches each character in the English word for each candidate character in the candidate lattice, and replaces the candidate order with the upper case or lower case.

(Effects of the Embodiment) Next, the effects of the above-described embodiment will be listed. First, in the language processing unit 106, when matching a candidate lattice and a dictionary, if there is an alphabetic character in the candidate character, uppercase letters or lowercase letters are included in the candidate lattice so that both uppercase and lowercase letters of the candidate character form a pair. By adding "," both uppercase and lowercase letters are always candidate characters, so it is possible to perform word matching even if the written description in the actual manuscript is not in the English word dictionary.

Therefore, there is no need to take measures such as increasing the number of entries depending on the notation of the English word dictionary.
There is no need to increase the dictionary size unnecessarily. In addition, since the dictionary search for English words can be performed without being aware of uppercase and lowercase letters, it is not necessary to newly prepare a dictionary matching algorithm dedicated to English words.

Secondly, as a result of the matching between the candidate lattice to which the candidate character is added and the word dictionary, the uppercase and lowercase letters of the alphabetic characters constituting the matched English word notation are replaced by the candidate rank of the recognized character. Since it is based on, the word notation described in the actual manuscript can be approximated and the recognition accuracy thereof can be improved.

Thirdly, as a result of matching the candidate lattice to which the candidate character has been added with the word dictionary, if the selected word is in all capital letters, the candidate word is evaluated based on the candidate rank of each recognized character. By determining whether or not the word should be the evaluation target in the unit 113, unnecessary words can be deleted from the evaluation target in the candidate word evaluation unit 113, for example, for abbreviations. Part 113
Processing accuracy is improved. In addition, this also reduces errors in linguistic analysis, which improves recognition accuracy.

Fourthly, as a result of collating the candidate lattice to which the candidate character has been added with the word dictionary, if the selected word is a mixed word of uppercase and lowercase letters, it is based on the candidate rank of each recognized character. By determining whether or not the word should be an evaluation target by the candidate word evaluation unit 113, unnecessary words can be deleted from the evaluation target of the candidate word evaluation unit 113 with respect to, for example, a place name or a person's name. Therefore, the processing accuracy of the candidate word evaluation unit 113 is improved. In addition, this also reduces errors in linguistic analysis, which improves recognition accuracy.

Fifth, by having an English word dictionary in which each word is described in one notation without describing all upper and lower case fluctuations in the word notation, the size of the word dictionary is not increased and the dictionary It is possible to save the trouble of creating a plurality of entries and reduce the work of creating an entry.

Sixth, the word dictionary matching algorithm is shared in both Japanese and English character strings, and a new dictionary matching algorithm dedicated to English words is not prepared.
For an OCR (optical character reader) that handles a mixed document of Japanese and English, it is possible to simplify and speed up the process.

[0045]

As described above, according to the character recognition device of the present invention (claim 1), when the candidate lattice is collated with the dictionary in the language processing means, if the candidate character has an alphabetic character. , By adding uppercase or lowercase letters in the candidate lattice so that both uppercase and lowercase letters become a pair, both uppercase and lowercase letters will always be candidate letters. Word matching can be performed even if it is not in the word dictionary.

Therefore, it is not necessary to increase the size of the dictionary because it is not necessary to increase the number of entries in the English dictionary. Also,
The dictionary search for English words can be done without regard to uppercase and lowercase letters, so there is no need to prepare a new dictionary matching algorithm for English words.

Further, according to the character recognition device of the present invention (claim 2), the collated English word notation is formed as a result of collation of the candidate lattice to which the candidate character is added and the word dictionary. Since uppercase and lowercase letters are replaced based on the candidate rank of the recognized characters, the word notation described in the actual manuscript can be approximated, and the recognition accuracy is improved.

Further, according to the character recognition device of the present invention (claim 3), when the selected word is all capitalized as a result of matching between the candidate lattice to which the candidate character is added and the word dictionary. Determines whether or not it is a word to be evaluated based on the candidate rank of each recognized character, so that unnecessary words can be deleted from the evaluation target for abbreviations, for example. The accuracy can be improved, linguistic analysis errors can be reduced, and the recognition accuracy can be improved.

Further, according to the character recognition device of the present invention (claim 4), as a result of matching between the candidate lattice to which the candidate character is added and the word dictionary, the notation of the selected word is mixed in upper case and lower case. If the word is a word, unnecessary words are deleted from the evaluation target, for example, for a place name or a person's name, by determining whether or not the word should be evaluated based on the candidate rank of each recognized character. As a result, processing accuracy is improved, linguistic analysis errors are reduced, and recognition accuracy is improved.

Further, according to the character recognition device of the present invention (claim 5), it has an English word dictionary in which each word is described in one notation without describing all the case fluctuations of the word notation. Therefore, the size of the word dictionary does not become large, and the effort of creating multiple dictionary entries is saved.
In addition, it is possible to reduce the amount of work for creating an entry.

According to the character recognition device of the present invention (claim 6), the word dictionary matching algorithm is shared for both Japanese and English character strings, and a new dictionary matching algorithm dedicated to English words is newly prepared. Since it is not necessary to do so, it is possible to simplify and speed up the processing in OCR that handles mixed documents of Japanese and English.

[Brief description of drawings]

FIG. 1 is a block diagram showing an overall configuration of a character recognition device according to an embodiment.

FIG. 2 is an explanatory diagram showing an example of a normal candidate lattice according to the embodiment.

FIG. 3 is a flowchart showing a processing operation of a language processing unit (post-processing unit) in FIG.

FIG. 4 is an explanatory diagram showing an example of a candidate lattice for English characters according to the embodiment.

FIG. 5 is an explanatory diagram showing an example of a candidate lattice for English characters after addition of uppercase and lowercase English letters according to the embodiment.

FIG. 6 is an explanatory diagram showing an example of an English word dictionary entry according to the embodiment.

FIG. 7 is an explanatory diagram showing an example of an English word determination unit according to the embodiment.

FIG. 8 is an explanatory diagram showing an example of an uppercase / lowercase correction unit according to the embodiment.

[Explanation of symbols]

 101 image input unit 104 character recognition processing unit 106 language processing unit (post-processing unit) 107 fixed character string output unit 108 candidate character lattice creation unit for English characters 109 uppercase / lowercase addition unit 110 word matching unit 111 English word determination unit 112 English capitalization・ Lower case correction part 113 Candidate word evaluation part

Claims (6)

[Claims] 1. A character recognition for recognizing a character string from image information, collating a recognition result character and a candidate lattice containing the candidate character with a word dictionary by a language processing means, and outputting a finally determined character string. In the apparatus, the language processing means creates an English character candidate lattice creation means for creating an English character candidate capable of containing a maximum of the candidate characters from the first candidate to the lower order for dictionary matching of English words, and a candidate. If a character has an alphabetic character, the uppercase and lowercase letters are added to the candidate lattice for English characters so that both uppercase and lowercase letters of the character are paired, and an uppercase and lowercase letter adding means is provided. Character recognizer. 2. The language processing means corrects the case of the notation of the selected word based on the candidate rank of each recognized character as a result of matching the candidate lattice to which the candidate character is added with the word dictionary. The character recognition device according to claim 1, further comprising an uppercase / lowercase correction means for performing the above. 3. The language processing means, based on a candidate rank of each recognized character, when the selected word is all capitalized as a result of matching between the candidate lattice to which the candidate character is added and the word dictionary. The character recognition device according to claim 1, further comprising an English word determination means for determining whether or not the word is to be evaluated. 4. The language processing means, as a result of collating the candidate lattice to which the candidate character is added with the word dictionary, when the selected word is a mixed word in uppercase and lowercase, a candidate for each recognized character. The character recognition device according to claim 1, further comprising an English word determination means for determining whether or not the word should be evaluated based on the rank. 5. The language processing means collates a candidate lattice with a word dictionary by an English word dictionary described in one notation for one word without describing all uppercase and lowercase fluctuations in the word notation. 5. Further, a word collating means for obtaining a candidate word is further provided.
Character recognition device according to any one of the above. 6. The character recognition device according to claim 1, wherein the language processing means shares a word dictionary matching algorithm for both Japanese and English character strings.