JPH0340079A - Post-processing method for character recognition in character reader - Google Patents

Abstract

PURPOSE:To obtain the result of recognition of higher accuracy by not only outputting the most coincident word by collating a word, but also outputting the information peculiar to its word or a pointer of a dictionary in which the intrinsic information is stored. CONSTITUTION:In the case code information peculiar to a word is added to every word stored in a word dictionary 4, and a word whose similarity is high is derived from combination of candidate characters by word collating means 2, 3, code information is outputted together, and based on the code information, necessary information is fetched. In such a way, the most coincident word is outputted by collating a word, furthermore, it is possible to cope with making character readers 1 - 5 intelligent since information peculiar to its word is obtained, thus, a result of recognition of high accuracy can be obtained.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a character recognition post-processing method for a character reading device,
In particular, the present invention relates to a character recognition post-processing method that improves recognition accuracy by matching recognition results read by a character recognition device in a character reading device with a word dictionary.

[Conventional technology]

Conventionally, as a method for improving character recognition accuracy of a character reading device, a character recognition post-processing method that performs matching with a word dictionary has been used. For example, JP-A-60-2174
The invention disclosed in Publication No. 90 improves recognition accuracy by selecting a plurality of virtual words from a string of character recognition results, determining their degree of similarity with dictionary words, and detecting the word with the highest degree of similarity. ing.

[IK problem that the invention attempts to solve]

In the above-mentioned conventional technology, only recognition of individual words can be expected. As character reading devices become more intelligent, you may want to recognize addresses in kana and convert them to kanji, or conversely, recognize addresses in kanji and output them in kana, or recognize addresses and add postal codes. do not have. Further, in general, much of the information written on a form is related to each other. For example, in addition to the address field, the postal code and telephone number may be entered, or furigana may be written. By using this information, it should be possible to further improve recognition accuracy. Incidentally, Japanese Unexamined Patent Publication No. 138478/1983 discloses an invention for checking an address using a postal code.

Conventional word matching techniques only output one or more candidates for the best matching word, and cannot be used for the above-described advanced information processing. As shown in FIGS. 9(a) and 9(b), in addition to the character recognition device 22, word matching device 23, and word dictionary 24 that make up the character reading device, a code conversion dictionary 26 is also required. and code conversion device 2
7 or check dictionary 29 or postal code check device z8
must be made in advance. As is clear from FIGS. 9(a) and 9(b), these dictionaries 26 and 29 must contain the same words as the words in the word dictionary, which is a waste of memory. Furthermore, in order to convert the code, it is necessary to search the conversion dictionary 26 for a word that matches the word matching result. In addition, Fig. 9 (a), (b)
21 indicates a form.

The present invention has been made in view of the problems of the prior art described above, and an object of the present invention is to provide a character recognition processing method in a character reading device that can eliminate memory waste and perform highly accurate recognition.

[Means to solve the problem]

The character recognition post-processing method in the character reading device of the present invention includes:
A character recognition means that reads a character string written on a form and outputs candidate characters, a word dictionary that stores a plurality of words, and combinations of candidate characters output from the recognition means and stored in the word dictionary. This is applied to a character reading device that includes a word matching means to find words with a high degree of similarity by matching words in the word dictionary. Code information unique to a word is added, and when words with high similarity are found from a combination of candidate characters by a word matching means, the above code information is also output, and necessary information is extracted based on the above code information. It is characterized by

- Code information includes (1) words related to the above words;
Or a group of words.

(2) An address or pointer where a word or word group related to the above word is stored.

(3) If words or word groups related to the above words are stored in the database, keywords or record numbers that can access them may be considered.

[For production]

According to the present invention, not only the most matching word is output by simple Hn matching, but also information specific to that word is obtained, so that it is compatible with the intelligentization of character reading devices and provides highly accurate recognition results. Obtainable. Further, it is also possible to output the obtained unique information together with the words or instead of the words, and it is possible to minimize the effort required to convert the recognition results into other information.

〔Example〕

The present invention will be described in more detail below with reference to embodiments shown in the accompanying drawings.

FIG. 1 is a block diagram showing an embodiment of the character reading device of the present invention. In FIG. 1, 1 is a form, 2 is a character recognition device, 3 is a post-processing device, and 4 is a word dictionary. The character recognition device 2 reads the characters written on the form 1 and outputs a plurality of candidate characters for each character to the post-processing device 3. Form 1
As shown in Figure 1, the following discussion will proceed assuming that the postal code (256) and address (Odawarafu) are entered.

In the character recognition device 2, since the postal code is a number, the recognition accuracy is high and it is recognized without any mistakes, and candidate characters are output one by one. Since addresses are written in Kanji, recognition accuracy is generally poor and many candidate characters are output. As shown in Figure 2, r for "small"
Three candidates are output: ``large'', ``small'', and ``r mountain'', and two candidates are output for ``田'', ``田'' and ``1 day'', and one candidate for ``hara'' is ``hara''. However, for “city”, “cloth” and “city”
Assume that two candidates are output.

The post-processing device 3 inputs these candidate characters from the character recognition device 2 and outputs the word with the highest degree of similarity as a candidate word. That is, the post-processing device 3 compares a word obtained by combining a plurality of candidate characters obtained corresponding to Odawara cloth with a word in the word dictionary 4 to find a match.

Assume that "Otawara-fu" is obtained as the first candidate, and "Odawara-fu" is obtained as the second candidate.

The word dictionary @4 has a structure as shown in FIG. In FIG. 3, reference numeral 6 is an address table, and the starting addresses in the word table 7 of a plurality of words starting with the first letter (for example, "小") of the candidate word can be found. In the address table 6, N indicates the number of words starting with "小", and Pl indicates the first address in the word table 7 of words starting with "small". Further, in FIG. 3, 8 is code information, and the words in the word table 7 and the code information 8 are each in a one-to-one correspondence. That is, word table 7
are sorted by first letter, and words that start with the same first letter are grouped together. The beginning of each group is
It can be obtained from address table 6. Further, in the case of FIG. 2, the code information 87 stores a postal code.

The second field of form 1 is read and word matched,
As mentioned above, if "Otawarafu" and "Odawarafu" are obtained as candidate words, the postal codes r324J and r256J are also output as their respective accompanying information based on the code information 8. If the first field reads r256J, you can check the postal code, exclude "Otawarafu" and select "Odawarafu".

FIG. 4 is a diagram showing another example of the word dictionary 4.

In FIG. 4, a pointer PI is stored as code information 8, and a pointer P points to a postal code stored in a postal code dictionary 9.

FIG. 5 is a diagram showing another example of the word dictionary 4.

In FIG. 5, a pointer P is provided corresponding to a word in the word table 7, and the pointer P indicates the data storage location of the postal code, prefecture name, and how to pronounce the same, which are stored in another database 10. In this case, the address table 6 shown in FIG. 3 may not be added.

In the above embodiment, the 8-check of the postal code is performed within the character reading device, but it is also possible to leave the check to a host device. In that case, the candidate word and code information are output as a pair. The host device can use the code information for checking or output it as data as is.

FIG. 6 shows a case where the reading is used as code data instead of the postal code. In this case as well, it can be used for checking furigana, etc., but it can also be used for converting from kanji to kana or from kana to kanji.

FIG. 7 shows a case where the record number ri of the database 11 outside the character reading device is output as the code information 8. It is assumed that the database 11 is composed of a large amount of information such as postal code, pronunciation, and prefecture name. When the record number rJ of the database 11 is output together with the word, the database 11 can be accessed to obtain the necessary information. In the example shown in FIG. 7, it is also possible to perform complex processing such as checking the postal code and furigana and outputting the name of the prefecture, or checking the postal code and outputting the pronunciation.

In addition, as code information 8, record number J4r r +
+7) You may also use a keyword that allows you to access the database.

FIG. 8 is a diagram showing another example of the word dictionary 4.

As shown in FIG. 8, the code information 8 stores, in addition to the pointer P2 of the dictionary 12, the number M of words in the related word group indicated by the pointer P2. Therefore, when a word with the highest degree of similarity is obtained through word matching, it becomes possible to generate a plurality of word groups related to that word. If the appearance frequencies of multiple word groups are known, it is possible to arrange them in order of appearance frequency, or if appearance frequency information 13 is stored as shown in Figure 8, it is possible to rearrange them in order of appearance frequency and output them. It is. Furthermore, if a pronunciation is entered in another field on the form 1, it can be used to check whether a word that matches the pronunciation is included in the word group.

〔Effect of the invention〕

As is clear from the above explanation, according to the present invention, not only the most matching word is found by word matching, but also the information unique to the word or the dictionary storing the unique information can be used. By outputting the pointer, more accurate recognition results can be obtained compared to information written in other fields. It is also possible to output the obtained unique information along with or instead of words, making it possible to respond to intelligent character reading devices and minimizing the effort of converting recognition results into other information. It is possible to limit

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the character reading device of the present invention, FIG. 2 is a diagram showing characters entered in a form and their candidate characters, and FIGS. 3 to 6 are words shown in FIG. 1. Figures 7 and 8 are diagrams showing an example of the structure of a dictionary; Figures 7 and 8 are diagrams showing an example of a combination of the word dictionary shown in Figure 1 and a database, etc.; Figure 9 (a);
, (b) is a block diagram showing an improved example of a character reading device according to the prior art. 1... Form, 2... Character recognition device, 3... Post-processing device, 4... Word dictionary, 5... Matching result, 6...
Address table, 7... Word table, 8... Code information, 9. Postal code dictionary, 10°11... Database, 12... Dictionary, 13... Appearance frequency information.

Claims (1)

[Claims] 1. Character recognition means for reading character strings written on a form and outputting candidate characters; a word dictionary storing a plurality of words; and character recognition means for reading candidate characters from the character string written on a form; In a character reading device, the character reading device includes word matching means for matching the combination with words stored in a word dictionary to find words with a high degree of similarity, for each word stored in the word dictionary, the word When a word with high similarity is found by a word matching means from a combination of candidate characters, the code information is also output, and necessary information is extracted based on the code information. A character recognition post-processing method in a character recognition device characterized by: