JPH0496190A - Device and method for post-processor for optical hand-written kanji - Google Patents

Abstract

PURPOSE:To improve the accuracy of recognition by extracting the reading of a corresponding KANJI (Chinese character) from a KANJI dictionary part with readings when the absence in a word dictionary part with FURIGANA (Japanese rendering of Chinese character) is judged from the result of the initial recognition and preparing a candidate character string corresponding to the reading based on that way of reading. CONSTITUTION:An optical initial recognition part 61 fetches a hand-written KANJI character string and the corresponding hand-written FURIGANA character string optically. Respective plural character candidates are taken in order of a constant level for each character forming each character string to output plural character string candidates arranged corresponding to each character string to be recognized as the result of the recognition. The retrieval is performed for each character string candidate which is the recognition result from a word recognition interpolation part 2 with HIRAGANA to a word dictionary 3 with FURIGANA, and when there is no each character string candidate, the reading is extracted from a KANJI dictionary 5 with reading for each word on each recognized KANJI, and the retrieval of the word dictionary 3 with FURIGANA is performed again while preparing the character string candidate corresponding to the reading. Thus, the accuracy of KANJI recognition can be improved.

Description

[Detailed Description of the Invention] [Table of Contents] Overview Industrial Application Fields Prior Art (Fig. 6) Means for Solving Problems to be Solved by the Invention (Figs. 1 and 2) Effects (Figs. 1 and 2)
Figure 2) Example (Figure 3.4.5) Effects of the invention [Summary] Optical system that searches a word dictionary with furigana that associates kanji with their furigana, and complements the optical recognition of handwritten kanji strings and furigana strings. The purpose of the present invention is to provide a post-processing device and method for optical handwritten kanji recognition that improves the accuracy of kanji recognition, realizes further efficiency in business processing, and is reliable. , optically captures a handwritten kanji string and a corresponding handwritten furigana string, lists a plurality of character candidates for each character forming each character string in a certain level order, and identifies each character string to be recognized for each level. Outputs multiple character string candidates arranged in correspondence with each other as recognition results,
The word dictionary with furigana is searched for each character string candidate that is the recognition result, and it is determined whether each character string candidate that is the recognition result exists in the word dictionary with furigana. outputs the character string candidate as a kanji with furigana, and if it does not exist, extracts the reading for each recognized kanji from a kanji dictionary with reading, and selects a string candidate that corresponds to the reading. This is a configuration in which the word dictionary with furigana is created and searched again.

[Industrial application field]

The present invention relates to a post-processing device and method for optical handwritten kanji recognition in an OCR device introduced in order to speed up business processing using forms, and in particular to a post-processing device and method for recognizing kanji and furigana to be added to the kanji. The present invention relates to a post-processing device and method for optical handwritten kanji recognition that searches a dictionary of words with furigana that are stored in correspondence and complements the optical recognition of a handwritten kanji string and a corresponding handwritten furigana string.

This increases the accuracy of Chinese character recognition by the OCR device, thereby making it possible to accurately read the values entered in the form, thereby realizing further efficiency in business processing.

[Conventional technology]

Conventionally, the post-processing of optical handwritten kanji recognition (particularly for addresses and names) carried out by various manufacturers is often carried out using word dictionaries with furigana.

An example of such a post-processing device for optical handwritten Chinese character recognition is shown in FIG.

As shown in the figure, this device includes a word dictionary with furigana section 63 that stores kanji strings and furigana strings to be attached to the kanji strings in correspondence, and a word dictionary section 63 that stores handwritten kanji strings and corresponding handwritten furigana strings. for each character forming each character string, and obtain the character string candidates with the highest level of similarity in correspondence with each character string to be recognized as recognition results. an optical initial recognition unit 61 for handwritten kanji with furigana that appears as It has the following.

[Problem to be solved by the invention]

By the way, the advantage of post-processing using a word dictionary with furigana is that if the furigana is reliably recognized and registered in the dictionary, even if the recognition rate on the kanji side is somewhat low,
It is in a place where you can recognize 0%. However, there are more than 100,000 names and address words in Japan, and in reality, only the most frequently used words are registered in a dictionary and used.

Also, although the recognition rate of furigana is higher than that of kanji,
Not 00%. Therefore, there have been problems in that the post-processing may not be effective due to the word dictionary not being registered in the dictionary for recognition supplementation using the word dictionary with furigana, or due to misrecognition of the furigana.

Such cases occur when the handwritten furigana is not registered in the dictionary, or even though it is originally registered, during the complementation process of the optical handwritten kanji recognition results using the word dictionary with furigana. This is a case where a part of handwritten furigana is incorrectly recognized and is determined to be unregistered.

Therefore, the present invention provides reliable optical post-processing for handwritten kanji recognition that can improve the recognition accuracy of kanji, accurately read values entered on forms, etc., and realize further efficiency in business processing. The purpose of this invention is to provide an apparatus and method.

[Means to solve the problem]

In order to solve the above technical problem, the first invention has a word dictionary with furigana part 3 that stores kanji and furigana to be added to the kanji in association with each other, as shown in FIG. In an optical handwritten kanji recognition post-processing device that searches the word dictionary with furigana section 3 and complements the optical recognition of handwritten kanji strings, the handwritten kanji strings and the corresponding handwritten furigana strings are optically captured and each For each character forming a character string, a plurality of character candidates are listed in order of level according to the degree of similarity, and for each level, a plurality of character string candidates are arranged corresponding to each character string to be recognized. An optical initial recognition unit l of handwritten kanji with furigana output as a recognition result, a word recognition complement unit 2 with furigana that searches for each character string candidate that is the recognition result for the word dictionary unit 3 with furigana, and kanji with furigana. The kanji dictionary with pronunciation unit 5 that stores the kanji and the reading of the kanji for each character in association with each other, and the word recognition complement unit 2 with furigana determine that the recognition result does not exist in the word dictionary with furigana unit 3. In this case, for each recognized kanji, the reading of the corresponding kanji is extracted from the kanji dictionary with reading 5, a character string candidate corresponding to the reading is created, and the character string candidate is sent to the word recognition complementation unit 2 with furigana. It has a complementing section 4.

On the other hand, the second invention, as shown in Figure 2, is an optical system that searches a word dictionary with furigana that associates kanji with furigana to be attached to the kanji, and complements the optical recognition of handwritten kanji strings. In the handwritten kanji recognition post-processing method, a handwritten kanji string and a corresponding handwritten furigana string are optically recognized, and for each character forming each character string, a plurality of character candidates are listed in order of level of similarity. , for each level, output a plurality of character string candidates arranged for each level in correspondence with the character strings to be recognized as recognition results (S1
), searches the word dictionary with furigana for each character string candidate that is the recognition result (S2), and determines whether each character string candidate that is the recognition result exists in the word dictionary with furigana. (S3), and if it is determined that the character string candidate exists, the character string candidate is used as a kanji with furigana (S5), and if it is determined that it does not exist, the character string candidate is The pronunciation is extracted from the kanji dictionary with pronunciation, a character string candidate corresponding to the pronunciation is created (S4), and the word dictionary with furigana is searched again (S2).

[Effect]

Next, the operation of the present invention (first and second inventions) will be explained.

As shown in FIGS. 1 and 2, the step Sl optically captures the handwritten kanji with furigana by the optical initial recognition unit 1 for handwritten kanji with furigana, and for each character forming each character, each Character candidates are listed independently in order of level according to the degree of similarity, and a plurality of character string candidates arranged for each level corresponding to each character string to be recognized are output as recognition results.

In step S2, for example, the furigana-added word recognition complementing unit 2 searches the furigana-added word dictionary unit 3 for each character string candidate that is the recognition result.

In step S3, the furigana-added word recognition complementing unit 2 determines whether or not there is a character string candidate that is the recognition result.

If any character string candidate exists as a result of the recognition, the process proceeds to step S5, and the character string candidate is output as the corresponding handwritten kanji with furigana.

On the other hand, if there are no character string candidates that are the recognition results in step S3, the process proceeds to step S4, and for example, the furigana complementation unit 4 extracts the characters from the reading kanji dictionary unit 5 for each recognized kanji. For each kanji character, the reading of the kanji is extracted, and based on the reading, it is determined whether the reading exists in the character candidates of the initial recognition result, and if it exists, the reading (furigana) is selected. The initial recognition result is corrected as correct, and the furigana-added kanji recognition complement section 2 searches the furigana-added word dictionary section 3 again in step S2.

If it is determined in step S3 that any character string candidate exists in the word dictionary section 3 with furigana, the process proceeds to step S5, and the corresponding character string candidate is output as a recognized handwritten kanji with furigana. become.

〔Example〕

Next, a post-processing device and method for optical handwritten Chinese character recognition according to an embodiment of the present invention will be described.

FIG. 3 shows a block diagram of the equipment configuration according to this embodiment.

As shown in FIG.
al Character Reader (optical character reading device) 30, a word dictionary with furigana unit 13 that stores kanji and furigana to be attached to the kanji in association with each other, a main storage device 40, an output unit 50, It has a kanji dictionary with readings 15 that stores kanji and readings of the kanji in correspondence with each character, and a kanji/furigana dictionary 60 that stores standard patterns of kanji and furigana for initial recognition. be.

FIG. 4 functionally shows the post-processing device for optical handwritten kanji recognition according to this embodiment, which includes a word dictionary section 13 with furigana, a string of handwritten kanji and the corresponding handwritten kanji. A furigana string is optically captured, and for each character forming each character string, a plurality of character candidates are independently listed in order of level according to the degree of similarity, and each character candidate is selected for each level in correspondence with each character string to be recognized. an optical initial recognition unit 11 for handwritten kanji with furigana that outputs a plurality of character string candidates arranged in a row as a recognition result, and the word dictionary unit 13 with furigana;
The recognition results are sent to the furigana-added kanji dictionary section 13 by the furigana-added word recognition complementation section 12 , which searches for each character string candidate that is the recognition result, the furigana-added kanji dictionary section 15 , and the furigana-added word recognition complementation section 12 . If it is determined that it does not exist, the reading of the corresponding kanji is extracted from the reading-reading kanji dictionary unit 15 for each recognized kanji, a character string candidate corresponding to the reading is created, and the word recognition with furigana is complemented. 12.

Here, the word recognition complementation unit 12 with furigana and the furigana complementation unit 14 correspond to the recognition completion post-processing unit 10, and are connected to the CPU 20, the main storage device 40 storing the processing program, and the output unit 50. It is equivalent.

Further, the OCR 30, the CPU 20, the kanji/furigana dictionary section 60, and the main storage device 40 storing the initial recognition program correspond to the optical initial recognition section 11.

Further, the optical initial recognition unit 11 for the handwritten kanji with furigana is configured as shown in FIG.
a and a handwritten furigana optical recognition section 11b.

Next, the operation of the optical handwritten kanji recognition post-processing apparatus and method according to this embodiment will be described.

The initial recognition results of the handwritten kanji string recognized by the handwritten kanji optical recognition section 11a of the handwritten kanji with furigana optical initial recognition section 11 and the handwritten furigana string recognized by the handwritten furigana optical recognition section 11b are The information is input to the word recognition complement section 12 of the recognition complement post-processing section 10 .

Here, the initial recognition of the handwritten kanji with furigana by the optical initial recognition unit 11 is performed by, for example, converting an optically captured handwritten character image into an electrical signal, and extracting only the character signal from the character signal containing noise. Cut out, normalize character position, size, inclination, line width, and extract features, and calculate the degree of similarity between the obtained pattern and the standard pattern stored in the kanji/furigana dictionary section 60 in advance. Then, a highly similar standard pattern is output as the initial recognition result.

As an example, FIG. 5 shows character string candidates as a result of initial recognition by the optical initial recognition unit 11 for the handwritten "bundle material (Azumamura)".

As shown in the figure, as a result of the initial recognition of the handwritten "Higashi", the candidate 1 with the highest level of similarity is "Higashi", and the candidate 2 with the next highest level of similarity is "Bundle". ”, and the next is the “strategy” of candidate 3, ..., candidate n
is "West". In addition, as a result of the initial recognition of the handwritten "village", candidate 1 with the highest similarity is "wood", the next candidate is "village", and the next candidate is "hayashi" (candidate 3).
...candidate n is "forest".

Similarly, for the handwritten "a", candidate 1 is "
Candidate 2 is "su", candidate 3 is "te", ..., candidate n is "fu". Thereafter, n candidates are listed for each of ``zu'', ``ma'', ``mu'', and ``wo'' written in the same way.

The furigana-added word recognition complement unit 12 searches the furigana-added word dictionary unit 13, and for each character string candidate, in the case shown in FIG. , Azumamuwo'', candidate 2 is ``Takuzai, Nasukohayo'',
. . , etc., it is determined whether the combination of each character string candidate exists in the furikana-added word dictionary section 13.

It is determined that none of the character string candidates arranged at the same level shown in FIG. 5 exist in the furikana-added word dictionary section 13.

On the other hand, if there is a furigana string for the kanji string in the furigana word dictionary section 13 and there is only one furigana string, and it matches the initial recognition result of the furigana, the furigana word recognition complement section 12 The furigana will be output as a recognition result.

As shown in FIG.
If it is determined that the name does not exist in 3, the furigana complementing unit 14 is notified of this fact.

Then, the furigana completion unit 14 searches for the reading of each kanji forming each initially recognized kanji string candidate, and searches the kanji dictionary with readings unit 15 for the corresponding reading among the initially recognized furigana combinations. A decision will be made as to whether or not.

In this case, if there is no corresponding one among the furigana combinations, the completion of the completion is reported to the furigana-added word recognition complementation unit 12 (no result).

In addition, if there is one corresponding combination of furigana among the initially recognized furigana combinations, if it is determined that there are multiple combinations, weighting is calculated for each candidate level to increase the probability. Make sure to choose something expensive.

In this way, the adopted result is fed back to the furigana-added word recognition complementation section 12 as a furigana complementation result.

Here, as an example searched using a kanji dictionary with readings,
Figure 5 shows the initial recognition results.
Regarding, "zai" does not fall under the furigana combination,
Regarding candidate 2, ``mura,''``Rin'' does not fall within the furigana combination, and ``Mura'' does fall within the furigana combination, so a predetermined probability value is calculated.

Furthermore, regarding Candidate 3, "Hayashi,""Rin" does not fall within the furigana combination, and "Hayashi" does not fall within the furigana combination. In this way, it is determined whether or not there is a corresponding furigana combination for candidate n, ``Mori''.

In this way, in the example shown in Figure 5, the judgment result is "uneven".
Instead, it is called "unevenness."

In this way, the furigana column candidates “Azuma” and “Mura” were obtained.
The initially recognized kanji candidate is searched again by the kanji recognition complementation unit 12 in the word dictionary with furigana unit 13, and if it exists in the dictionary unit 13, the final recognition result is: The kanji string and furigana string are output from the output section 50.

As explained above, according to this embodiment, the accuracy of kanji recognition by optical reading by an OCR device is improved, and the burden on the operator involved in correcting input data caused by the low recognition accuracy of conventional methods is reduced. becomes possible.

〔Effect of the invention〕

As explained above, in the present invention, when the initial recognition result is determined by the furigana-added word recognition complementing unit and the furigana-added word dictionary unit to be non-existent in the furigana-added word dictionary unit, each recognized kanji is Extract the reading of the corresponding kanji from the kanji dictionary with reading, and based on the reading,
A furigana completion section is provided to create character string candidates corresponding to the reading.

Therefore, the accuracy of kanji recognition by optical reading is improved,
After reliable and efficient optical handwritten kanji recognition, it is possible to reduce the burden on the operator who had to correct the input data due to low accuracy as in the past. Processing apparatus and methods can then be provided.

[Brief Description of the Drawings] Fig. 1 is a principle block diagram of the first invention, Fig. 2 is a principle flow diagram of the second invention, Fig. 3 is a device configuration block diagram of the embodiment, and Fig. 4 is a principle block diagram of the second invention. FIG. 5 is a block diagram of the embodiment, FIG. 5 is a diagram showing a processing example of the embodiment, and FIG. 6 is a block diagram of a conventional example. 1.11... Optical initial recognition unit for handwritten kanji with furigana 2.12... Word recognition complement unit with furigana 3.13...
・Word dictionary with furigana section 4.14... Furigana complement section 5.15... Kanji with reading blade - dictionary section

Claims (2)

[Claims] (1) It has a word dictionary with furigana (3) that stores kanji and the furigana to be attached to the kanji in association with each other, and searches the word dictionary with furigana (3) to find handwritten kanji strings. In an optical handwritten kanji recognition post-processing device that complements optical recognition, a handwritten kanji string and a corresponding handwritten furigana string are optically captured, and multiple character candidates are created similar to each character for each character forming each character string. An optical initial recognition unit for handwritten kanji with furigana ( 1
), a word dictionary with furigana part (3), a word recognition complement part with furigana (2) that searches each character string candidate that is the recognition result, and a word dictionary with furigana part (2) that searches for each character string candidate that is the recognition result; If it is determined by the kanji dictionary with pronunciation (5) and the word recognition complement unit (2) with reading that the recognition result does not exist in the word dictionary with furigana (3), the recognition For each kanji that has been read, extracts the reading of the corresponding kanji from the kanji dictionary with reading (5), creates a character string candidate corresponding to the reading, and processes the word recognition complement unit (2) with furigana.
A post-processing device for optical handwritten kanji recognition, comprising: a furigana complementing unit (4) that sends a reading to a user. (2) In a post-processing method for optical handwritten kanji recognition, which complements optical recognition of handwritten kanji strings by searching a word dictionary with furigana that associates kanji with furigana to be added to the kanji, and the corresponding handwritten furigana string, and for each character forming each character string, list a plurality of character candidates in order of level according to the degree of similarity, and select each of the above-mentioned characters to be recognized for each level. A plurality of character string candidates arranged in correspondence with the character string are output as recognition results (S1), and a word dictionary with furigana is searched for each character string candidate that is the recognition result (S2). It is determined whether each resulting character string candidate exists in the word dictionary with furigana (S3), and if it is determined that it exists, the character string candidate is output as a kanji with furigana (S5); If it is determined that it does not exist, for each recognized kanji, the reading is extracted from the kanji dictionary with readings, a character string candidate corresponding to the reading is created (S4), and the reading is read again. A post-processing method for optical handwritten kanji recognition, characterized by searching a kanji dictionary with attached words (S2).