JPH0340434B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a character recognition post-processing method, in particular, to a pattern-recognized candidate character, focusing on whether or not there is a specific character among the candidate characters. This invention relates to a character recognition post-processing method in which post-processing is performed by finding a break in a character string and comparing the extracted character string with a dictionary.

[Conventional technology and problems]

FIG. 3 shows a diagram for explaining problems with the conventional method.

Conventionally, in character recognition devices that recognize kanji using optical means, for example, in order to increase the recognition rate,
For example, post-processing is performed to check against a specific knowledge dictionary such as an address dictionary. According to the conventional method, as shown in Figure 3, when entering an address on the input sheet 1, post-processing is required unless the character strings are entered separately for each prefecture, city, town, and village. The problem was that I couldn't do it. That is, in order to perform post-processing, it was necessary to use a formatted paper that had input boxes divided in advance into prefecture input frames, city/town/ward input frames, and ward/town/village input frames on input sheet 1. .

As a result, the number of input boxes increases, making it difficult to format the paper, making it difficult to fill in information, and making it difficult to read written characters.

[Means for solving problems]

The present invention aims to solve the above-mentioned problems, and performs post-recognition processing on character strings written in solid form, for example, in the case of addresses, character strings written without separating the prefecture, city, town, etc. To provide a character recognition post-processing method that improves the rate. Therefore, in the character recognition post-processing method of the present invention, candidate characters are output for each character as a result of recognition of an input character string, and a character recognition device that performs post-processing to select a final candidate for these candidate characters. In the character recognition post-processing method, there is a specific character registration section that registers and stores specific characters that delimit the input character string in advance, and a level determined by the specific characters for the character strings delimited by the specific characters. a dictionary that stores meaningful terms; a specific character search unit that searches the candidate characters to find the position of the specific character among the candidate characters; and position information of the specific character obtained by the specific character search unit. The present invention is characterized by comprising a candidate character string extraction section that extracts a character string from the candidate characters based on the above, and a dictionary matching section that compares the character string extracted by the candidate character string extraction section with the dictionary. .

[Effect]

The present invention utilizes the fact that a specific input character exists among candidate characters as a result of recognizing input characters,
For example, in the case of an address, it sequentially searches for the presence of specific pre-registered characters such as prefecture, city, county, ward, town, etc. in the candidate characters, and separates the input string at the specific character position. This makes it possible to perform post-recognition processing on solid character strings.

〔Example〕

Hereinafter, embodiments will be described with reference to the drawings.

FIG. 1 shows the configuration of an embodiment of the present invention, and FIG. 2 is a diagram for explaining the processing aspect of the embodiment of the present invention.

In the figure, 1 is an input sheet, 2 is a character input unit such as OCR, 3 is a recognition unit that analyzes patterns of input characters and selects candidate characters, 4 is a candidate memory in which the selected candidate character strings are stored, and 5 6 is an address dictionary, 7 is a specific character search unit that searches for specific characters in the candidate memory 4, and 8 is a specific character registration section in which specific characters that separate character strings are registered in advance. A candidate character string extraction unit extracts a candidate character string; 9 is a dictionary collation unit that collates the candidate character string extracted by the candidate character string extraction unit 8 with the address dictionary 6; 10 is a result output unit that outputs a dictionary collation result represents. Also, in Figure 2,
Reference numeral 20 represents an input character string.

In the case of the present invention, addresses can be written in solid writing on the input sheet 1 without dividing them into prefectures, cities, towns, etc., for example, as shown in the input character string 20 shown in FIG. 2. The character input section 2 is, for example,
Input sheet 1 by optical means such as OCR
The light intensity information is transmitted to the recognition unit 3.
The recognition unit 3 selects candidate characters for the input characters by, for example, extracting topological features or performing stroke analysis on the input information, and various methods are well known for this recognition process. The detailed explanation will be omitted.

The recognition unit 3 ranks each of the selected candidate characters in descending order of so-called dissimilarity as shown in FIG. 2, and stores them in the candidate memory 4 from the 1st to the 20th rank. Needless to say, the first candidate character is not necessarily the correct input character.

By the way, when the input character string is an address, the address usually includes the following characters:
This will include specific characters such as "mura". The specific character registration section 5 stores the levels of these specific characters according to the order in which they appear in the input character string.
It is registered and stored in advance. In the case of an address, for example, the kanji for ``都, 道, ふ, 県'' are registered at the prefecture level, the kanji for ``city, county, ward'' are registered as the city/town/ward level, and the kanji for ``ku, town, village'' are registered at the city/town/ward level. ” are registered at the ward, town, and village level. Further, in the specific character registration section 5, index information for a dictionary storing character strings that can appear at that level, etc. is set corresponding to each level or each specific character.

When the candidate character string is stored in the candidate memory 4, the specific character search unit 7 refers to the specific character registration unit 5,
Using a specific character corresponding to each level as a key, whether or not the specific character serving as the key appears among the candidate characters is sequentially checked from the beginning. In the example shown in Figure 2, "Miyako" is placed in the first position of the third character.
You will find the characters. As a result, it can be seen that the first to third characters are character strings at the prefecture level. Next is the 4th
By checking sequentially from the 6th character, "city" will appear in the 6th character, so from the 4th character to the 6th character.
The first three characters are recognized as being at the city/town/ward level. Similarly, it is recognized that the 7th to 10th characters are words at the ward, town, and village level.

Note that these specific characters do not necessarily have to appear first in the candidate ranking; for example, for locations where specific characters are likely to appear, such as the third or fourth character, the candidates are ranked first, starting with the highest candidate ranking. Search results are selected in order.

When the specific character search unit 7 detects the position of the specific character, the candidate character string extraction unit 8 extracts partial candidate characters from the candidate character string stored in the candidate memory 4 until the specific character appears. Cut out the columns,
The dictionary checking unit 9 is notified.

When the notified character string is, for example, at the prefecture level, the dictionary matching unit 9 compares each word registered in the dictionary A part of the address dictionary 6 with the characters according to the candidate ranking in the notified partial candidate character string. It is sequentially checked to see if the combinations match. As a result, in the case of the example shown in FIG. 2, it can be seen that the word is "Tokyo" at the prefecture level. In addition,
If there are two or more matching words in the comparison with the address dictionary 6, by calculating the candidate ranking points,
The character combination with the higher candidate ranking is selected.

Similarly, at the city/town/ward level in the example shown in FIG. Obtained as a matching result.
Furthermore, at the ward/town/village level, the 7th to 10th letters
"Shinkoji Town" is determined by the partial candidate character string up to the first character.
is obtained as the matching result.

The verification results are notified to the result output unit 10, and the result output unit 10 confirms with the inputter as necessary, determines the final recognition result, and outputs it to a predetermined device, etc. Execute.

The above explanation has been given using the character input of an address as an example, but the present invention is generally useful for separating character strings, for example, when inputting affiliation in a company, "department" or "section" is used as a specific character. The same can be applied to anything that has characters that appear.
Furthermore, the present invention is applicable not only to handwritten characters but also to recognition of printed characters.

〔Effect of the invention〕

As explained above, according to the present invention, a solid input character string can be divided into word units that can be post-processed, so post-processing is performed to select the most appropriate final candidate from a plurality of candidates. It is possible to improve the recognition rate. Since the input character string can be written in solid form, it is easy to write and the entered character string is easy to read. Additionally, paper waste can be reduced. The inputter does not need to be aware of specific characters, and there is no burden on the inputter.

[Brief explanation of drawings]

Fig. 1 shows the configuration of an embodiment of the present invention, Fig. 2 is a diagram for explaining the processing mode of an embodiment of the present invention, and Fig. 3 is a diagram for explaining problems with the conventional method. . In the figure, 1 is an input sheet, 2 is a character input section, 3 is a recognition section, 4 is a candidate memory, 5 is a specific character registration section,
6 is an address dictionary, 7 is a specific character search section, 8 is a candidate character string extraction section, 9 is a dictionary collation section, 10 is a result output section, and 20 is an input character string.

Claims (1)

[Claims] 1 In a character recognition post-processing method in a character recognition device that outputs candidate characters for each character as a recognition result of an input character string, and performs post-processing to select the final candidate for those candidate characters, the above input characters are a specific character registration unit that registers and stores specific characters that delimit columns; a dictionary that stores meaningful terms corresponding to levels determined by the specific characters for character strings delimited by the specific characters; a specific character search unit that searches the candidate characters and locates the position of the specific character among the candidate characters;
A candidate character string extraction unit that extracts a character string from the candidate characters based on the position information of the specific character obtained by the specific character search unit, and the above-mentioned character string extracted by the candidate character string extraction unit. A character recognition post-processing method characterized by comprising a dictionary collation unit that collates with a dictionary.