JPH0652367A - Post-processing method for character recognition result - Google Patents

Abstract

PURPOSE:To provide a post-processing method for a character recognition result by which an accurate recognition result can be obtained in word unit even when an erroneous recognition result is obtained in character unit, and for which small amount of information to be generated in advance is required, and which can easily conform to the change of the information to be generated in advance. CONSTITUTION:A character is classified based on the error trend of the character. No information relating to character classification is stored in a word dictionary. A reference word for the reference of retrieval is formed from the character recognition result, and classification name notation is generated from the reference word based on the character classification, and it is confirmed whether or not a word with the same classification name notation as that of the classification name notation exists based on the character classification when the word is retrieved. A candidate word is narrowed down from the word with the same classification name notation as that of the reference word.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a post-processing method for character recognition results, and more particularly to improving recognition performance by searching for words in consideration of error tendency of the recognition results.

[0002]

2. Description of the Related Art For example, as a machine translation system, there is a machine translation system in which a character recognition device is applied to an input means to simplify an input operation by a user. For example, in the character recognizing device applied in this way, of course character-by-character recognition,
Word-by-word recognition is also important. In the past, a method has been already proposed in which a correct recognition result can be obtained for each word even if there is an erroneous recognition result in recognition for each character.

For example, reference is made to “Numakura et al.,“ Information Retrieval System Retrievable with Incorrect Key ””, Journal of Information Processing Society of Japan, Vol. 30, No. 11, pp. 1468-1478, November 1989, for example. be able to.

Hereinafter, a word search method (post-processing of character recognition results) according to the method disclosed in the above-mentioned document, in which an accurate recognition result of a recognition target word is obtained from the recognition results of respective characters constituting the recognition target word The process which constitutes a part of the method) will be described.

When applying this method, it is necessary to prepare in advance a character classification and a hierarchical word dictionary classified by the notation of each word. Here, the character classification is a classification of all characters into some classes based on the error tendency of the characters, and each class is given a class name. The word class name notation is a notation formed by arranging the class names to which the characters that make up the word belong.

When recognizing a word, first, a class name notation X is created by replacing the recognition result of each character forming the recognition target word with the class name of the character classification. Next, the class name notation X having the highest degree of coincidence with the class name notation X thus obtained is searched from the class name notation of the above-mentioned word dictionary. Finally, the word in the above-mentioned word dictionary having the class name notation Y is set as a search target, the word having the highest degree of coincidence with the above-mentioned recognition result is searched, and the word is set as the search result.

[0007]

By the way, the error tendency of each character described above differs depending on the font. Therefore, the method of classifying each character differs depending on the font. If only the classification is recreated, a complicated work is not required, but according to the above-mentioned conventional method, the word dictionary also needs to be recreated, and a very complicated work is required.

Such a problem occurs even when the font is not changed. For example, even if the classification of each character is reviewed and changed by learning, even if the class to which a certain character belongs is changed, the class name notation of the word dictionary is widely changed. Therefore, such a thing is not practical and, in the end, limits the improvement of word recognition accuracy.

In practice, researches on character-by-character recognition methods have been actively conducted so as to improve the recognition accuracy on a character-by-character basis. When such a new character recognition method is applied, the error tendency may be different from the conventional one even if the font is the same, and the above-mentioned problem also occurs in such a case.

The present invention has been made in consideration of the above points, and in recognition in character units, an accurate recognition result can be obtained in word units even if there is an erroneous recognition result. Therefore, it is an object of the present invention to provide a post-processing method for a character recognition result that requires a small amount of information to be created in advance and can easily cope with a change in information created in advance.

[0011]

In order to solve such a problem, in the present invention, a process of searching a word dictionary from a recognition result of each character forming a recognition target word to obtain a candidate word for the recognition target word is performed. In the post-processing method of the character recognition result including, all the characters are classified into some classes based on the error tendency of the characters, the class names are given and stored in advance, and the class names are stored in the word dictionary. Words are stored without accompanying class name notation, and from the recognition result of each character,
A first process of creating a reference word for reference at the time of searching, a second process of creating a class name notation in which the class names of the classes to which each character of this reference word belongs are stored, and a second process stored in the word dictionary. Existing words as a search target, it is determined based on the character classification whether the class name notation matches the reference word class name notation,
And a third process of excluding the different words as not satisfying the requirements of the candidate words.

[0012] Here, among the words not excluded in the third processing, it is preferable that the candidate word has the smallest number of characters that do not match the reference word.

[0013]

In order to make it possible to easily cope with a change in the character classification, it is sufficient to omit the class name notation of the word dictionary, which requires a great change in accordance with the change in the character classification. The present invention has been made according to such an idea.

That is, even if the character classification is prepared, the word dictionary is not accompanied by the notation of class names. Then, a reference word for search reference is formed from the character recognition result, a class name notation is created based on the character classification, and at the time of word search, it is checked whether there is a word having the same class name notation as this class name notation. It was decided to confirm based on the character classification.

Words in the word dictionary with different class names are not candidate words, but even if the class names are the same, that alone is not sufficient for narrowing down candidate words. Therefore, it is preferable that the candidate name is the one in which the class name notation is the same as that of the reference word and the number of characters that do not match the reference word is the smallest.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A post-processing method for character recognition results according to the present invention will be described in detail below with reference to the drawings regarding an embodiment in which it is applied to recognition of English words.

Although not shown in the drawings, this embodiment is practically realized in hardware, for example, by a workstation equipped with an optical character reader (OCR). The functional block is as shown in FIG.

In FIG. 2, the document reading means 11 reads the document written on the paper as binary data. The character recognition unit 12 cuts out each character area, each word area, and the like based on the binary data, obtains the characteristic amount of the binary pattern of each character area, and uses the reference value created in advance for each character. The recognition result of each character is obtained according to the distance from the feature amount, and the recognition result storage unit 13 is provided for each word area.
To be stored in. The reference word creating means 14 creates a reference word to be referred to at the time of search based on the recognition result for each character. The character classification storage unit 15 stores a character classification in which all the characters are classified based on the error tendency of each character, together with a class name (classification name). The contents stored in the character classification storage means 15 will be described later. The word dictionary 16 stores at least English words that appear to appear in general sentences, and is not hierarchized by class name notation. In the case of this embodiment, the search speed is intended to be shortened and fixed for each number of characters, and even within the same number of characters, the hash values are fixed for each and stored. As the hash value, for example, the remainder obtained by dividing the sum of the ASCII code of each character in the word by 64 is applied. Candidate word search means 1
7 is a word that has a class name notation consisting of the class names of the characters that make up the reference word and has a class name notation that matches this class name notation The dictionary 16 is searched, and the searched candidate words are temporarily stored in the recognition result storage means 13. The output word determination means 18 determines an output word from the candidate words by using the initial recognition result, and the result output means 2
1 to print out or display.

The process flow of the embodiment method implemented by such functional blocks is as shown in FIGS.

First, a large processing flow will be described with reference to FIG.

A character area, a word area or the like is cut out from the document data converted into binary data, a characteristic amount of a binary pattern (character pattern) of each character area is obtained, and a reference value is created in advance for each character. The recognition result of each character is obtained according to the distance from the feature amount, and the recognition result of each character is arranged for each word area (step 100). FIG. 4 shows an example of the character recognition result, and is an example in which the input recognition target word (correctly, its pattern) is “take”. Here, even if the character to be read this time and the character at the time of creating the reference feature amount are the same character, the feature amount differs due to the difference in the character pattern such as the font, so the distance is not always the minimum. I won't. Here, all the characters whose distance is less than or equal to a certain threshold value are extracted as the recognition result.

When the recognition result is obtained for each character forming the word in this way, a reference word is formed by connecting the first candidate characters for each character (step 101).

FIG. 5 is a chart showing the information obtained as the process steps progress. As shown in FIG.
When the recognition result as shown in is obtained, the reference word is "t.
ahe ”.

When such a reference word is obtained, the stored character classification is used to obtain a class name notation, and the word dictionary is searched using this class name notation as a key to obtain candidate words (step 102). ). The processing of this step is characteristic of this embodiment, and the details will be described later.

When one or more candidate words are obtained in this way, the word to be output is determined using the recognition result obtained in step 100 (step 103). For this output word determination processing, for example, the method described in Japanese Patent Application No. 3-196509 and drawings can be used.
That is, the reference feature amount of each character of the candidate word is used to obtain the distance from each character of the recognition target word, and the sum of the distances obtained for each character is used as the evaluation value of this candidate word. Then, the candidate word having the smallest evaluation value is set as the word to be output.

Then, the determined output word is output by printing or displaying and a series of processing is completed (step 1).
04).

Next, the word dictionary search process in step 102 will be described in detail with reference to FIG.

When the reference word is created and the word dictionary search process is started, first, as shown in FIG. 1A, the hash value of the reference word is calculated and the hash value is set in the parameter hash (step 200). The calculation of the hash value is of course the same as the method of calculating the hash value of the word dictionary, and as described above, it is, for example, the remainder obtained by dividing the sum of the ASCII code of each character in the reference word by 64.

Next, a class name notation for the reference word is created (step 201). At this time, the classification of the character created in advance based on the error tendency is used. Figure 6
Shows a storage example of classifications, and is a table in which class names are associated with characters belonging to the classes. In such a classification, reading errors occur in the same class, but reading errors do not occur in characters of other classes. FIG. 6 shows, for example, the letter “a”,
This means that a read error may occur between [o], "u", and "v". The method described in the above document can be applied to such character classification. When the classification shown in FIG. 6 is applied to the reference word “tahe” as shown in FIG. 5 described above, “DABC” is obtained as the class name notation, as shown in FIG.

After that, when the matching process of the words in the word dictionary is repeated for each word as described later, the number of characters that do not match the reference word is a parameter for setting the smallest value at that time. The number of non-matching characters is set to a large value which is practically unthinkable as the minimum number of non-matching characters (step 202). As will be described later, the candidate word is
Select from the number of characters that do not match the reference word as much as possible.

After the above initialization processing is completed,
Proceed to the candidate word determination process by matching each word in the word dictionary. In such candidate word determination processing, first, the hash value is equal to the hash value of the reference word and the number of characters (word length) is equal to the number of characters (word length) of the reference word. Is checked (loop consisting of steps 203 to 206), and then it is checked whether each word in the word dictionary whose hash value is different from the hash value of the reference word but whose number of characters is equal to the number of characters of the reference word is a candidate word. (Step 20)
3 to 209 loop).

The hash value is equal to the hash value of the reference word and the number of characters (word length) is the number of characters of the reference word (word length)
The loop processing for collating each word in the word dictionary that is equal to is a candidate word is detailed in FIG. 1B for collating the word with the processing for extracting the word satisfying the above condition from the word dictionary (step 203). Processing (step 20)
4), the process of confirming whether the target word matches the reference word (step 205), and the process of confirming whether the process for all the words satisfying the above conditions has been completed (step 206). Become.

Here, in the loop processing in which each word in the word dictionary whose hash value is equal to the hash value of the reference word and whose number of characters is equal to the number of characters of the reference word is a candidate word, it is not matched with the reference word. The number of characters is 0
If it finds one, that is, if the reference word itself is in the word dictionary, the dictionary search is immediately terminated there (Yes in step 205).

On the other hand, a loop process for checking whether each word in the word dictionary whose hash value is different from the hash value of the reference word but whose number of characters is equal to the number of characters of the reference word is a candidate word is a word that satisfies this condition. The process of extracting from the dictionary (step 207) and collating the word are shown in FIG.
It is composed of a process shown in detail in (B) (step 208) and a process of confirming whether the process for all the words satisfying the condition has been completed (step 209).

As described above, the collating order of the words in the word dictionary is changed according to the hash value in order to shorten the search time when the reference word itself correctly recognizes the word to be read. Is. Although recognition is erroneous, it is often recognized correctly, and by defining the search order in this way, the overall search processing time and thus the recognition processing time can be shortened. .

For any word in the word dictionary whose hash value is equal to the hash value of the reference word and whose number of characters is equal to the number of characters of the reference word, the hash value is different from the hash value of the reference word, but the number of characters is equal to the number of characters of the reference word. The matching process (steps 204 and 208) for each word is the same regardless of the word in the word dictionary, and the details are as shown in FIG. 1 (B).

When a word in the word dictionary is to be matched, after setting a parameter diffcount for counting the number of non-matching characters of the word to the reference word to an initial value 0,
Steps 301 to 301 for each character of the matching target word
The loop consisting of 05 is repeated.

First, with respect to the character extracted in step 301, it is confirmed whether or not the class name of the character matches the one in the same order of class name notation for the reference word (step 302). If a negative result is obtained here, the collation for this word is immediately terminated and the main routine (step 20
5 or 209). On the other hand, if it is the same as the classification in the class name notation, it is determined whether or not the character itself matches the character in that order in the reference word (step 30).
3). If they are different, the parameter diffcount is incremented by 1, and if they are the same, it is immediately confirmed whether or not the last character has been processed (steps 304, 30).
5).

In this way, when the confirmation of the class name with respect to the last character and the confirmation of the match / mismatch with the characters of the reference word are completed, the number of mismatched characters with the reference word in the matching target word, that is, the value of the parameter diffcount and The size is compared with the smallest minimum number of non-matching characters in the matched word (step 306). If the number of non-matching characters (value of diffcount) in this matching target word is larger than the minimum number of non-matching characters, this matching target word is not added to the candidate words and the main routine (step 205
Or, return to 209). When the number of non-matching characters (value of diffcount) in the matching target word this time is equal to the minimum number of non-matching characters, this matching target word is added to the candidate words and the process returns to the main routine (step 205 or 209) (step 307). If the number of non-matching characters (value of diffcount) in the matching target word this time is less than the minimum number of non-matching characters, the candidate word so far is discarded, this matching target word is registered as the candidate word, and the minimum number of non-matching characters is Main routine (step 205 or 209) by replacing with the number of unmatched characters (value of diffcount) in the matching target word
(Steps 308 and 309).

By executing the processing shown in FIG. 1 described in detail above, when the reference word is in the word dictionary, the word becomes a candidate word, and when the reference word is not in the word dictionary, it is similar to the reference word. One or more words stored in the word dictionary having the same name notation and the smallest number of characters that do not match the reference word are equal to the reference word are candidate words.

When the reference word is "tahe" and its class name is "DABC" as shown in FIG. 5 and the character classification shown in FIG. 6 is applied, the hash value is the reference word. A candidate word cannot be obtained by the matching process for each word in the word dictionary that is equal to the hash value and the word length is equal to the word length of the reference word, but the hash value is different from the hash value of the reference word, but the word length is the reference word. The matching process for each word in the word dictionary equal to the word length of
e ”can be obtained as a candidate word.

Therefore, according to the above embodiment, if the characters are classified according to the error tendency, that is, without classifying each word for each class name notation, the candidate word for the recognition target word is selected. Obtainable.

As a result, even if the font of the document to be read changes from the one used up to now and the error tendency changes, it is possible to deal with it by changing only the character classification without changing the word dictionary. Further, even when the character classification is changed by learning, the existing word dictionary can be applied as it is. Furthermore, even if the error tendency changes by changing the applied character recognition method, it can be dealt with by changing only the character classification without changing the word dictionary.

In the above embodiment, the case where an English word is a recognition target has been described, but the present invention is also applicable when a word in another language is a recognition target. For example, in the case of Japanese, the extraction of words and the recognition of words are performed in parallel, but the present invention can be applied.

In the above embodiment, one reference word is set to obtain the candidate word, but two or more reference words are set from the character recognition result to obtain the candidate word. good.

Further, in the above embodiment, whether or not the reference word is found in the word dictionary is distinguished from the search of other words having the same hash value as that of the reference word. It may be possible to first determine whether or not it is in the word dictionary. In such a case, the word classification may not be necessary based on the hash value.

Furthermore, in the above-described embodiment, the word dictionary stored in the word dictionary is compared with the reference word for the same name by comparing each character unit. It is also possible to obtain the class name notation of the word in the above by using the character classification, and then compare them.

Further, in the above-mentioned embodiment, the condition of the candidate word is that the number of characters that do not match the reference word is the smallest. However, the condition for the number of characters that do not match may be set somewhat looser than this.

[0049]

As described above, according to the present invention, only the character classification information is prepared without including the information about the class name notation in the word dictionary. Since the word can be obtained, the amount of information to be created in advance is small, and a post-processing method for character recognition results that can easily cope with changes in information created in advance can be realized.

[Brief description of drawings]

FIG. 1 is a flowchart showing a word dictionary search process according to an embodiment.

FIG. 2 is a functional block diagram showing a configuration for realizing an embodiment.

FIG. 3 is a flowchart showing a series of character recognition processing including processing according to an embodiment.

FIG. 4 is an explanatory diagram showing an example of a recognition result.

FIG. 5 is an explanatory diagram showing information obtained by the processing of the embodiment.

FIG. 6 is an explanatory diagram showing an example of character classification according to the embodiment.

[Explanation of symbols]

15 ... Character classification storage means, 16 ... Word dictionary, 17 ... Candidate word search means, 302 ... Processing step for determining match between word class name notation in reference word dictionary and reference word class name notation.

Claims (2)

[Claims] 1. A post-processing method for character recognition results, which includes a process of searching a word dictionary from the recognition results of each character that constitutes a recognition target word to obtain candidate words for the recognition target word, based on the error tendency of characters. Then, all the characters are classified into several classes, class names are given and stored in advance, and words are stored in the word dictionary without class name notation in which the class names are arranged. , A first process of creating a reference word to be referred to at the time of search from the recognition result of each character, and a second process of creating a class name notation in which the class names of the classes to which each character of this reference word belongs are arranged Then, the words stored in the word dictionary are searched, and it is determined whether the class name notation matches the class name notation of the reference word based on the character classification. The third process to exclude as not satisfying and Post-processing method of the character recognition result, characterized in that it comprises. 2. The character recognition result according to claim 1, wherein among the words not excluded in the third processing, a candidate word has the smallest number of characters that do not match the reference word. Post-processing method.