JPH0944604A - Character recognizing processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To narrow down comparative collation object and to shorten the time for comparative collation by selecting words including the same number of key characters among a lot of words even when a lot of words having the same number of characters exist. SOLUTION: The key character having the low error rate of recognized result is detected out of candidate characters. When such a key character is detected, a dictionary filter part 4 is generated for narrowing down the word of collation object when reading this word out of a word dictionary 3. The number of candidate character strings is counted from the word dictionary 3 and the word having the same number of characters as that is extracted. In this case, that kind becomes (e) or (s) as a result of key character detection. Besides, concerning (e), the number becomes two characters and concerning (s), it becomes one character. Then, the word having five characters, two key characters of (e) and one key character of (s) is selected from among a lot of word groups stored in the word dictionary 3. Thus, the number of words to be the collation object is decreased.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition processing method for automatically correcting a recognition result when a document image is read to extract and recognize a character by using a word dictionary.

[0002]

2. Description of the Related Art An apparatus for reading handwritten characters and printed characters as image data and cutting out the image in character units for recognition processing automatically inputs data such as a document into an information processing apparatus or performs various arithmetic operations. Widely used to perform processing. Such a character recognition device recognizes the pattern character by character and compares it with a dictionary prepared in advance to perform recognition processing of each character. However, handwritten characters and the like do not always exactly match the patterns of the dictionary, and some recognition errors occur. For example, it may not be possible to determine whether the character is t or l as a result of the recognition process of the character t. In such a case, first, one character or two or more candidate characters are listed for each character. Then, the character string forming one word is collated with the word dictionary. From the word dictionary, words having the same number of characters as the number of characters forming the character string are extracted and compared one by one. The word with the highest matching rate is taken as the character recognition result, and the correct reading rate of character recognition is improved. Such a post-processing technique is described in, for example, Japanese Patent Publication No. 61-20038.

[0003]

The character recognition processing method as described above has the following problems to be solved. A large number of words that will be output as a result of the character recognition processing are stored in the word dictionary in advance. Therefore, when the number of words registered in the dictionary is large, a large number of words having the same number of characters exist in the dictionary, and it takes a long time to match the words. This causes a problem that the speed for character recognition processing becomes slow. The present invention has been made in view of the above points, sets a character with a low error rate of the recognition result as a key character, narrows down the words to be matched by focusing on the number of key characters and the position of the key character, The purpose is to shorten the time for comparison and matching.

[0004]

[Means for Solving the Problems]

<Structure> The character recognition processing method of the present invention reads an image of a document, cuts out the image for each character and performs recognition processing, and converts a word formed by a candidate character string obtained as a result of the recognition processing into a word. In the case where post-processing of the recognition result is performed by comparing and collating with a word group composed of the same number of character strings extracted from the dictionary, a key character with a low error rate of the recognition result is set in advance and the key character is included. The number of key characters of the word composed of the candidate character string is counted, and the word group composed of the same number of character strings extracted from the word dictionary and including the same number of key characters is compared and collated. <Operation> Even if there are many words with the same number of characters, the comparison and collation target is narrowed down by selecting the word including the same number of key characters.

<Structure> Another character recognition processing method of the present invention reads an image of a document, cuts out the image in character units, performs recognition processing, and is composed of candidate character strings obtained as a result of the recognition processing. A word that compares and collates a word group composed of the same number of character strings extracted from the word dictionary and performs post-processing of the recognition result, in advance, sets a key character with a low error rate of the recognition result, For a word composed of a candidate character string containing the key character, a word composed of the same number of character strings extracted from the word dictionary by detecting the position of the key character and containing the key character corresponding to the same position Compare and collate with group. <Operation> If words having the same number of key characters in the same position are selected from a group of words having the same number of characters, the words to be compared and collated can be further narrowed down. The above means can also be applied to the processing of a character string obtained by voice recognition. In any case, it is preferable to select a key character that has a high appearance rate in any word as much as possible.

[0006]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be described in detail below with reference to the embodiments shown in the drawings. <Narrowing Down by Number of Key Characters> FIG. 1 is an explanatory diagram for explaining the outline of the character recognition processing method of the present invention. First, the outline of the present invention will be described with reference to this figure. The input document 1 describes a document that is a target of character recognition. Here, as an example, one word "reset" is shown. When recognizing each character that makes up a word,
First, the input document 1 is read as image data, and characters are cut out character by character from the image data.
As a result, the five characters shown in FIGS. 2-1 to 2-5 are cut out as constituting this word, and the recognition process by pattern recognition or the like is performed for each character. Here, when such recognition processing is performed, one or several candidate characters to be output as a recognition result are detected. Here, candidate characters S and V are obtained for the third character 2-3, and l and t are obtained for the fifth character 2-5.

In the present invention, a key character having a low recognition result error rate is detected from such candidate characters.
The key characters vary depending on the character recognition processing method and the characteristics of the recognition processing device. Therefore, for example, depending on the device, the letter e shown in the figure may have a very low error rate in the recognition result, and the letter a may have a low recognition rate. Therefore, such key characters are arbitrarily set for each device. Since the key character is set in order to increase the character recognition rate, it is preferable to set the character that appears relatively frequently in each word as the key character. Therefore, even if the recognition result has a low error rate, characters that do not appear in too many words do not necessarily need to be set as key characters.

When such a key character is detected, the dictionary filter unit 4 for narrowing down the matching target word is read out when the matching target word is read out from the word dictionary 3 shown in the figure. From the word dictionary 3, the number of candidate character strings is counted, and words having the same number of characters as that are extracted. This is because when character recognition is performed on one word, it is considered that an error in the number of characters is extremely rare. Here, the word "re
Since it is a process for "set", in the conventional case, all the words of five characters are taken out from the word dictionary 3. Here, as a result of detecting the key character, the type is e or s.
Becomes Also, the number is two characters for e and one character for s. Therefore, a word having five characters, two key characters e and one key character s is selected from a large number of word groups stored in the word dictionary 3. As a result, the number of words to be matched is reduced, and "reset" is output as the post-processing result. In another embodiment described later, the number of words is further narrowed down in consideration of the position of the key character.

FIG. 2 shows a hardware block diagram for implementing the present invention. In order to carry out the present invention, as shown in the figure, a character recognition unit 6, a key character table 7, a key character detection unit 8, a word dictionary 3, a dictionary filter unit 4, a word matching unit 9, an output unit 10, and the like are provided. Use the equipment provided. The character recognition unit 6 is a unit that reads an image of an input document, performs character extraction and character recognition described in FIG. 1, and outputs a character code of a single character or a plurality of candidate characters for each character. The structure of this part is exactly the same as that of the conventional device. The word matching unit 9 receives the recognition result from the character recognition unit 6 on a word-by-word basis, and compares and matches the words one by one with the words registered in the word dictionary 3 extracted through the dictionary filter unit 4 to obtain the highest matching score. This is a part where a high word is selected as a candidate word. For the calculation of the degree of coincidence, various well-known methods such as the number of coincident characters, the degree of coincidence in pattern comparison or the rank of candidate characters can be used.

The key character detection unit 8 receives the recognition result for each word from the character recognition unit 6, and how many key characters are registered in the key character table 7 in the words output as the recognition result. This is the part that counts. In addition,
As described above, the key character table 7 stores and displays characters e and s, which are preset key characters. This key character table can be rewritten, for example, every time the character recognition method of the character recognition unit 6 is changed. The dictionary filter unit 4 pays attention to the number of key characters output by the key character detection unit 8 and the key characters, and the word dictionary 3
This is a part for selecting a word to be matched among the words registered in and outputting it to the word matching unit 9. The word dictionary 3 is configured so that the dictionary filter unit 4 can quickly select a word based on the output of the key character detection unit 8. This will be described later with reference to FIG.

The output unit 10 is a unit that determines a character string for outputting the post-processing result based on the candidate words output from the word matching unit 9. This part has the same structure as the post-processing device after the conventional character recognition process. FIG. 3 shows an explanatory diagram of an example of a word dictionary. As described above, when the candidate characters are set to e and s, for example, the word dictionary 3 is configured to efficiently extract the corresponding word based on the number of key characters and the number of characters forming the word. The example shown in FIG. 3 is a table in which only a portion in which words having a word length of 5 characters including key characters e and s are collected is extracted and tabulated. Such a table can be easily retrieved by classifying the word dictionary according to the number of key characters included in the word. For example, in this figure, the number of key characters e is "0" and the number of key characters s is "1".
It can be seen that the words are "first" and "burst" .... In the example shown in FIG. 1, the number of key characters e is “2” and the number of key characters s is “1”. Therefore, from the word dictionary shown in FIG. 3, “size” and “rese”
Two words "t" are extracted.

The key character has a high recognition rate, but it may be erroneously recognized. Therefore, for example, if the key character e is 2 characters, the key character s is 0 to 2 characters, and the key character e is 1 to 3 characters, the key character s
You may make it extract a candidate word from a certain range, such as "1". Even with this, compared to the case where all the words with the same number of characters are extracted,
It is possible to sufficiently narrow down the words to be compared and matched. In the example shown in FIG. 1, the candidate character is the second character 2-2 and 5
There are two characters for each of the second characters 2-5. Therefore, character strings formed by such combinations of candidate characters are "resel", "revel", "rese".
There are four types, “t” and “revet”. When these are compared and collated with the words extracted from the dictionary filter unit 4, “reset” is output as a candidate word.

As a result, among the words extracted from the word dictionary, the words to be compared and matched are sufficiently narrowed down, and the processing speed for the matching processing is increased. Further, the possibility that a candidate word in which the key character portion is replaced with another character is selected is reduced, and the correct reading rate is improved. For example, if matching is performed with all words having a word length of 5 characters, two words "revel" and "reset" are output as candidate words from the dictionary shown in FIG. A constant matching rate is calculated for each of the candidate characters during pattern recognition. Therefore, any one of the characters is output as the first candidate with a high matching rate, and the other characters are output as the second candidates. However, as a result of such recognition processing, the “rev
In many cases, "el" is determined as the final candidate word. However, when the key characters e and s are used for narrowing down the collation target by using the present invention, “revel” does not become a candidate word, so that misrecognition does not occur.

<Refining by Key Character Position> In the above embodiment, an example of narrowing down the words to be collated by the number of key characters has been described. However, the key character with a low error rate of the recognition result has high reliability not only in the number but also in the position in the word. Therefore, in the next embodiment, narrowing down is performed by focusing on the position of the key character. For this purpose, the key character detection unit 8 shown in FIG. 2 is configured to accept the recognition result output by the character recognition unit 6 in word units, detect the key character position, and output the key character position to the dictionary filter unit 4.
Further, the dictionary filter unit 4 focuses on the key character position, takes out a word having the same number of characters from the word dictionary 3 and having a key character at the corresponding position, and sends it to the word matching unit 9.

FIG. 4 shows an example explanatory diagram of a word dictionary used in such an embodiment. In this figure, key characters are e and s, and words are arranged corresponding to the character positions. That is, of the words consisting of 5 characters, the words in which the key character e is arranged from the 1st to the 5th are listed, respectively. "0" is a word for which the key character e does not exist. The same applies to the key character s. Again, the operation will be specifically described using the example of FIG.

The candidate character obtained as a result of the recognition processing in the example of FIG. 1 includes the key character e in the second character 2-2 and the fourth character 2-4 as described above. The key character s is included in the third character 2-3. The detection result is sent to the dictionary filter unit 4. Here, the dictionary filter unit 4 extracts from the word dictionary shown in FIG. 4 the words whose key character e is at the second and fourth positions and whose key character s is at the third character position. The result is "resin", "reset" and "ousel" shown in the figure.
It will be four words "laser". That is, there are a great number of words having the same number of characters as shown in this figure, but if the focus is placed on the position of the key character, the collation target is narrowed down in this way. In the word dictionary shown in FIG. 4, words having the same key character at a plurality of positions are redundantly registered at several places. For example, in the word "reset", the key character e exists in the second and fourth key,
May be registered in both places. Therefore, words may be extracted in duplicate. In this case, either one is extracted.

As described above, as in the case of using the number of key characters, the words to be compared and collated are effectively narrowed down. Note that the words may be narrowed down by focusing on both the number of key characters and the positions of the key characters.

The present invention is not limited to the above embodiment. In the above embodiment, the post-processing using the result of character recognition was described.However, in a device that converts various arbitrary data such as voice recognition into a character code string, a code string to be output is a dictionary. When making a comparison, the present invention can be similarly adopted.

[Brief description of drawings]

FIG. 1 is an explanatory diagram of a character recognition processing method of the present invention.

FIG. 2 is a hardware block diagram for implementing the present invention.

FIG. 3 is an explanatory diagram of a word dictionary example (No. 1).

FIG. 4 is an explanatory diagram of a word dictionary example (No. 2).

[Explanation of symbols]

 1 Input Document 2-1 to 2-5 Characters to be recognized 3 Word dictionary 4 Dictionary filter unit

Front page continued (72) Inventor Makoto Torikoshi 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd.

Claims (4)

[Claims] 1. A document image is read, the image is cut out in character units for recognition processing, and a word composed of candidate character strings obtained as a result of the recognition processing is extracted from a word dictionary with the same number of characters. In the case where post-processing of recognition results is performed by comparing and collating with a group of words, a key character with a low error rate of the recognition result is set in advance, and it is composed of a candidate character string containing that key character. A character recognition processing method characterized by counting the number of key characters of a word, and comparing and matching with a word group that is composed of the same number of character strings extracted from a word dictionary and that contains the same number of key characters. 2. An image of a document is read, the image is cut out in character units for recognition processing, and a word composed of candidate character strings obtained as a result of the recognition processing is extracted from a word dictionary with the same number of characters. In the case where post-processing of recognition results is performed by comparing and collating with a group of words, a key character with a low error rate of the recognition result is set in advance, and it is composed of a candidate character string containing that key character. Character recognition characterized by detecting the position of a key character for a word and comparing and matching it with a word group that consists of the same number of character strings extracted from the word dictionary and that contains the key character corresponding to the same position. Processing method. 3. The character recognition processing method according to claim 1, wherein the result of recognition processing in character units instead of the image of the document is compared and collated with a word group. 4. The character recognition processing method according to claim 1, wherein only characters having a high appearance rate are set as key characters.