JPH06215198A - Character recognition post-processing system - Google Patents

Abstract

PURPOSE:To simultaneously handle Japanese characters and alphabetical letter strings and to enable collation in the unit of meaning as well by combining the unit of collation with the unit of a word in the case of character recognition post-processing. CONSTITUTION:With the character recognized result 20 as a key character, a dictionary reading processing 11 reads character arrangement information from a dictionary 21. In dictionary preparation processing 12, the read charcters arrangement information is analyzed, the partition of character is detected, and partition information or the character arrangement information is divided to a prescribed length. At a collation part 13, the character recognized result 20 is collated with the divided character arrangement information. In the case of collation, the partitioning character is specially collated. The result of the character arrangement information is stored and defined as the collating result with one piece of character arrangement information of the dictionary.

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,
In particular, it relates to a post-processing method of character recognition for confirming / correcting the reading result.

[0002]

Post-processing of character recognition is widely used as a means of assisting incomplete character recognition.

As for the nature of the input character string, in the US and Europe, the character string is delimited in units of words, whereas the Japanese character string is customarily written as a solid character. Therefore, the post-processing method differs depending on the processing target. ing.

In post-processing for Western texts, the character recognition result is divided into word units based on delimiter information such as blanks, and collation is performed in word units. In addition, in the processing for Japanese solid writing, a character string divided into delimiter information such as white space is also collated without assuming that it is a single word.

[0005]

When the above-mentioned character recognition post-processing method is applied to mixed reading of Japanese and Western sentences, it may be necessary to write Japanese in Japanese or Western, depending on the method adopted. However, there is a problem in that there may occur a case where data is continuously extracted without delimiter information.

[0006] Further, like the title of a book, there is a character string that is a sequence of words in terms of structure, but is better to handle as a single word in terms of meaning. When dealing with such a character string in the post-processing of the conventional method described above, such a character string is considered to be a sequence of words and registered in the dictionary for each word, or the sequence of words itself is set as There is a method of registering as a word in the dictionary. However, it is difficult for the former to be regarded as one in terms of meaning, and for the latter, there is a problem that the processing time becomes enormous because the word length becomes long.

[0007] Further, by knowing a group of words in the meaning, it is possible to express the relationship between them (for example, a book name and its author's name) with a simple structure. There is a problem in that it is possible to handle only a set of word strings, or it is not sufficient to classify word strings in semantic units.

[0008] Furthermore, in English, uppercase letters ← → depending on the format.
It is necessary to deal with such fluctuations because there are often changes in lowercase letters and omitting names with initials. There is a problem that the processing becomes very complicated if the matching corresponding to such a variation is performed without any contradiction at the time of matching.

The present invention has been made in view of the above-mentioned conventional circumstances, and therefore an object of the present invention is to provide a novel character recognition post-processing method capable of solving the above-mentioned problems inherent in the prior art. To provide.

[0010]

In order to achieve the above object, the character recognition post-processing method according to the present invention is a means for reading out information to be collated from a dictionary according to a result of character recognition, and a character recognition result. And means for generating a hierarchical division dictionary by dividing the read character string according to the read character arrangement information, and means for collating the input character string with the read character arrangement information and sequentially obtaining the similarity in hierarchical order. And means for finally determining the post-processing result corresponding to the input character string.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described in detail with reference to the preferred embodiments thereof with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention (the invention according to claim 1). The outline of the processing flow is as follows.

Referring to FIG. 1, when the character recognition result 20 is given, the process start 10 activates the dictionary reading 11.

FIG. 2 is a diagram showing an example of the character recognition result 20, in which three candidates of the character recognition result are output for each written character.

The dictionary reading 11 searches the dictionary 21 by using the character recognition result 20 as a key character, selects and reads word information to be collated. The dictionary generation unit 12 sets the collation range of the input character string centering on the break information, analyzes the read word information based on this information, and sequentially generates collation information divided into short parts as necessary. , And the matching positions are sequentially passed to the matching unit 13.

After all the word information is finally transmitted to the collating unit 13, the information of the end of transmission is finally transmitted to the collating unit 13 and the next dictionary reading is executed.

When the word division information does not match the input character string break information, the dictionary generation is stopped and the next dictionary reading is executed.

The collating unit 13 collates the division information of the character sequence and the character recognition result 20 which are sequentially transmitted, and accumulates the result. When the collation unit 13 receives the transmission end information from the dictionary generator 12, the collation unit 13 compares the collation intermediate result 22 with the cumulative result, and collates the result when it is considered that there is a possibility of remaining in the final determination result. Add to intermediate result 22.

When there are no more dictionaries to be read in the dictionary reading 11, the judgment 14 is activated, the judgment result 23 is created by referring to the collation intermediate result 22, and the processing is ended.

The character recognition post-processing system having the above-described structure is composed of a CPU (central processing unit) and a storage medium (RAM, hard disk, floppy disk, etc.) for storing various processing programs and data. Can be configured.

The processes of processes 11 to 13 will be described in detail below.

In the dictionary reading process 11, the character recognition result 20
Is used as a key character, and character arrangement information is read from the dictionary 21. An example of the character arrangement information is shown in FIG. FIG. 3 shows one character arrangement information divided into semantic units, the character storage position shows the storage position of the character code string to be referred to, and the number of characters is the number of character code to be referred to in that area. Is shown. The concatenation information indicates the connection relation with the character arrangement information of another semantic unit, and the content information indicates the information with meaning about this information. Further, Δ in the character code string indicates a blank code.

In the dictionary generation processing 12, the character recognition result 20
Search for a delimiter and record its position. Two types of delimiters are extracted: blank and other symbols (“,”, “;”). Next, the delimiter is searched from the read character arrangement information, the delimiter position is compared with the delimiter position of the character recognition result, and when they match, the character arrangement information is divided. If there is no delimiter, the character arrangement information is divided into a predetermined length. The comparison of delimiters targets all delimiters except the last character, and only blanks are compared in the last character.

FIG. 4 shows an example in which the character arrangement information of FIG. 3 is divided from the character recognition result of FIG.

The divided character arrangement information is sequentially transmitted to the collating unit 13, and after the final information is transmitted, the information of the transmission completion is transmitted.

In the comparison of the delimiter characters described above, the delimiter information of the character sequence information and the delimiter information position of the character recognition result match, and the information other than the delimiter of the character sequence information and the information other than the blank space of the character recognition result. It will be carried out by matching.

The collating unit 13 will be described with reference to FIG.

Referring to FIG. 5, the collation unit 13 calculates the similarity between the divided character arrangement information and the character recognition result 20. The similarity calculation 132 confirms whether or not there is a candidate character designated by the character arrangement information at the designated position of the character recognition result. For example, the candidate rank of character recognition may be used as the similarity value. However, in the case of delimiter characters, use the delimiter character table registered in advance,
The best delimiter character candidate is detected and replaced with the character division information, but this detection is not included in the similarity calculation.

When all the characters except the delimiter of the divided character arrangement information are not present at the corresponding positions in the character recognition result, the reject flag 135 is set. In other cases, the similarity is accumulated 133 and stored in the primary storage buffer 136 together with the collation position and the character arrangement information.

When the information indicating the end of transmission of the division information is received, the collation candidate is added 134. First, the reject flag 134 is checked. If the reject flag 134 is not set, the primary storage buffer 136 is compared with the collation intermediate result 22, and if there is a possibility that it remains as a final judgment candidate, it is added to the collation intermediate result 22 and the collation intermediate result 22 is added. If the number of 22 candidates exceeds a predetermined number, the most unlikely candidate is deleted.

FIG. 6 is a processing flowchart showing the character recognition post-processing method according to the second aspect. The dictionary reading 31 performs a dictionary search based on the matching intermediate result 22 in accordance with the dictionary search using the candidate character of the character recognition result 20 as a key character. At this time, the dictionary search information is used as the connection information in FIG.

In the judgment 32, the judgment result 23 is created by combining the verification candidates from the verification intermediate result 22 based on the connection information and the verification position.

According to the third aspect of the present invention, in the dictionary generation of the first or second aspect, when it is detected that the divided character arrangement information is composed of alphabets or alphabets including numbers, three types of uppercase ← → lowercase conversion are performed. Create character arrangement information. FIG. 7 shows the division information created from FIG.

The invention described in claim 4 is the same as in claim 1,
In the dictionary generation of No. 3, when the divided character arrangement information is composed of alphabets, the number of divisions is two or more, and the surname and first name are described in the content information of the character arrangement information, the other than the final division information is replaced with the initials. Is. FIG. 8 shows an example of this generation.

[0035]

As described above, according to the present invention,
It is said that the character sequence information in which Japanese notation and alphabet notation are mixed can be handled in the same semantic unit, and that the processing speed can be processed at the same level as Japanese even for alphabets that generally have longer characters than Japanese notation. The effect is obtained.

Further, in the invention according to the second aspect, it is possible to perform collation based on a concatenation relationship in a semantic unit such as a book name and an author name.

The invention according to claim 3 can cope with the instability of writing uppercase / lowercase letters in alphabetical notation with one registration information, and can also cope with the case where an English sentence is included in a part of the Japanese sentence. Is.

In addition to the invention described in claim 4, 1
With one registration information, it is possible to respond to changes in the entry of the family name.

Therefore, the present invention has a great effect as a post-processing of character recognition for an entry form having contents represented by entry of book name-author name-publisher name.

[Brief description of drawings]

FIG. 1 is a functional block configuration diagram showing an embodiment of the invention described in claim 1.

FIG. 2 is a diagram showing an example of a character recognition result.

FIG. 3 is a diagram showing an example of a storage format of one character arrangement information in a dictionary.

FIG. 4 is a diagram showing divided character arrangement information generated based on the character arrangement information of FIG.

5 is a block diagram showing a detailed function of a matching function 13. FIG.

FIG. 6 is a functional block diagram showing an embodiment of the invention described in claim 2.

FIG. 7 is a diagram showing character arrangement information generated from the character arrangement information of FIG. 3 according to the invention described in claim 3;

FIG. 8 is a diagram showing an example of generation of divided character arrangement information according to the invention described in claim 4;

[Explanation of symbols]

 10 ... Start of processing 11, 31 ... Dictionary reading processing 12 ... Dictionary generation processing 13 ... Collation processing 14, 32 ... Judgment processing 15 ... End of processing 20 ... Character recognition result 21 ... Dictionary 22 ... Collation intermediate result 23 ... Judgment result 131 ... Division End 132 ... Similarity calculation 133 ... Result accumulation 134 ... Collation candidate addition 135 ... Reject flag 136 ... Primary storage buffer

Claims (4)

[Claims] 1. A character recognition post-processing method for confirming / correcting the result of character recognition based on a dictionary in which information about the arrangement of characters is written, and information to be collated is read from the dictionary according to the result of character recognition. Means, a means for dividing the character string according to the character recognition result and the read character arrangement information to create hierarchical divided character arrangement information, and the input character string and the read character arrangement information are collated, and their similarity A character recognition post-processing method comprising: a means for sequentially obtaining the post-processing in a hierarchical order; and a means for finally determining a post-processing result corresponding to an input character string. 2. In a character recognition post-processing method for confirming / correcting the result of character recognition based on a dictionary describing information on the arrangement of characters, information to be collated is read from the dictionary according to the result of character recognition. Means, a means for dividing the character string according to the character recognition result and the read character arrangement information to create hierarchical divided character arrangement information, and the input character string and the read character arrangement information are collated, and their similarity And a means for finally determining the post-processing result corresponding to the input character string, and selecting the dictionary information to be matched and reading the dictionary, the input character string and the already read A post-processing method for character recognition, which is characterized in that the result related to collation of dictionary information is used as a read condition together with the result of character recognition. 3. When reading a dictionary according to a character recognition result, when the character recognition result includes alphabets, the dictionary is read without distinguishing between uppercase and lowercase, and when the divided dictionary is created, the divided dictionary includes only alphabets and numbers. Generates three types of split dictionaries: a split dictionary that consists only of uppercase letters, a split dictionary that consists of only lowercase letters, and a dictionary that consists of only uppercase letters and lowercase letters. The character recognition post-processing method according to claim 1 or 2, further characterized in that the dictionary having the highest similarity is selected in advance. 4. When creating a divided dictionary, the divided dictionary is composed of only alphabets and numbers, and when the content of the dictionary information is limited to first and last names, a dictionary composed of initials is generated. 2. The character recognition post-processing method according to any one of 2 and 3.