JPS59158482A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve the accuracy of character reading by providing the 1st and 2nd recognizing sections and discriminating and outputting the final answer based on a preset statistical data for recognition from each result of recognition. CONSTITUTION:Suppose that an input pattern is (a), the result of recognition R1 of an ANK recognizing section 2 is (a) and the result of recognition R2 of a Kanji (Chinese character) recognizing section 3 is ryo. A discriminating circuit 4 references a discriminating table and reads reference data SA, SB of an address designated by an answer (a) of the result R1. Further, the data SA, SB are compared with the value of similarity of the result R1 and the difference of similarity of the result R2 respectively. In this case, when the value of similarity is 90 and the difference of the similarity is 20, since the values exceed the threshold values SA, SB (80, 15), the circuit 4 discriminates the (a) being the answer of the result R1 of the recognizing section 2 as the final answer and outputs it. When the values of the results R1, R2 are smaller than the data SA, SB respectively, the recognizing section 3 outputs ryo being the answer of the result R2 as the final answer.

Description

[Detailed description of the invention] [Technical field of invention] The present invention relates to an optical character reading device.

[Technical background of the invention and its problems]

In recent years, optical character reading devices (hereinafter referred to as OCR) are capable of reading texts that include kanji (including hiragana) in addition to alphanumeric characters, kana characters, and special symbols (collectively referred to as kanji). There is a kanji OCR that can do it. In such Kanji OCR, since the characters to be read are ANK characters (approximately 128 character types) and Kanji characters (more than 2000 character types), the number of recognized character types is large. Therefore, unlike a character recognition method using only ANK characters, this method is complicated and large-scale, as the dictionary memory has a large capacity.

However, while conventional Kanji OCR can read a large number of character types, the recognition accuracy is lower than when recognizing ANK characters included in a sentence or when using a character recognition method for ANK characters. There were drawbacks. This is because the number of character types to be recognized is wide and large, and the recognition accuracy per character is lowered.

[Purpose of the invention]

This invention was made in view of the above circumstances, and its purpose is to improve the accuracy of character reading by making it possible to reliably recognize lin characters even when reading texts consisting of melon characters and kanji. The object of the present invention is to provide a character recognition device that can perform the following tasks.

[Summary of the invention]

This invention provides both an ANK recognition section that recognizes ANK characters made up of a limited range of character types and a Kanji recognition section that recognizes characters made up of a wide range of character types. The recognition results of both recognition units are given to a recognition matching determination unit, and this determination unit determines and outputs the final answer. In this case, the recognition matching determination unit stores in advance statistical data for recognition based on differences in similarity values as a determination table, and outputs a final answer based on this statistical data.

This makes it possible to reliably recognize ANK characters in the process of reading sentences made up of ANK characters and Chinese characters.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a character recognition system according to the present invention. In the figure, reference numeral 1 denotes a preprocessing circuit, into which a character i4 turn P obtained by scanning a form is input, and performs preprocessing such as noise removal on the character pattern P. The character pattern P usually consists of a photoelectrically converted digital signal. The character Tsukutan P output from the preprocessing circuit 1 is
It is given to both the ANK recognition unit 2 and the kanji recognition unit 3. ANK Recognition Recognizes a limited range of characters (ANK characters) such as alphanumeric characters, kana characters, and special symbols.

The kanji recognition unit 3 recognizes characters in sentences consisting of ANK characters and kanji (including hiragana). Recognition matching judgment circuit (
(hereinafter referred to as a determination circuit) 4 is composed of a microprocessor, etc., and the recognition results R1, which are output from both recognition units 2 and 3,
For R2, recognition is determined based on statistical data for recognition stored in memory as a determination table in advance, and a final answer is output.

The operation of such a configuration will be described with reference to FIGS. 2 and 3. Assume now that characters recorded on a form 10 as shown in FIG. 2 are to be recognized. First, the reading field 11h consisting of only ANK characters of the form 10 is scanned, and a character pattern P is given to the melon recognition section 2 and the kanji recognition section 3 for each normal character.

In this case, the character pattern P is a digital signal that has been photoelectrically converted as described above, and passes through the preprocessing circuit 1 to both the recognition units 2 and 2.
given to 3. The recognition results R1 and R2 of both recognition units 2 and 3 are then given to a determination circuit 4. In this case, the determination circuit 4 determines and outputs the ANK recognition recognition result R for the character pattern P in the reading field lla of the form 1o based on format control data prepared in advance. This format control data is normally stored in the memory of an OCR control device (not shown), and is data indicating the line data content for each reading field for the form 1o to be read.

Next, assume that a sentence consisting of ANK characters, Chinese characters, etc. is read in the reading field 11b of the form 10. In this case as well, the character pattern P for each character is ANK
Recognition Recognition 2 and Kanji - 5 = given to both recognition units 30. The determination circuit 4 obtains recognition results R1 and R2 from both recognition units 2 and 3. here,
Each recognition result R1, R2 is not just an answer, but also the 1st to Nth candidate characters in each recognition result R1, R2, the 1st similarity value, and the 1st and 2nd candidates. It consists of data on similarity differences between characters, etc. The determination circuit 4 outputs an answer without using a determination table if the following conditions are satisfied. That is, each answer k1. of the recognition results R1, R2. If A2 are the same, the answer A2 from the kanji recognition unit 3 is output as the final answer. In this case, when both recognition results R1 and R2 are rejected, the determination circuit 4 determines that the recognition results are rejected. Furthermore, if the answers Al and A2 are not the same and the recognition result R1 is rejected, the answer A2 is set as the final answer. If the answer A1 does not exist among the first to Nth candidate characters of the recognition result R2, then the answer A2 is set as the final answer.

Under such conditions, the determination circuit 46- determines that each recognition unit 2.3 has a different answer kl, 12 recognition results R.
1 and R2 respectively, and the answer A of the recognition result R1.
1 exists among the 1st to Nth candidate characters in the recognition result R2 (i.e., the recognition result of the kanji recognition unit 3), refer to the determination table (statistical data for recognition) as described above. It turns out. This determination table is configured, for example, as shown in FIG. 3, and the address is designated by A on the ANK recognition section. The reference data specified by this address is the threshold of 8 people and the kanji recognition unit 3 for comparison with the similarity value of the first candidate character picked up by the ANK recognition recognition unit.
It consists of a threshold value SR for comparing the similarity difference between the first and second candidate characters obtained in . Specifically, for example, if the input character (turn P is "A"), the ANK recognition recognition unit's knowledge result R1 is "A", and the recognition result R2 of the kanji recognition unit 3 is "A".
Suppose that the result is 1 completion.

If "A" is found among the first to Nth candidate characters in the recognition result R2, the determination circuit 4 refers to the determination table as described above. That is, the determination circuit 4 outputs the reference data sA and SR of the address specified by the answer "A" of the recognition result R1 in the table of FIG. Then, the reference data s people and SB are compared with the similarity value of the recognition result R1 and the similarity difference of the recognition result R2, respectively. In this case, if the similarity value of R) is 90 and the similarity difference of R2 is "20", the determination circuit 4 sets the threshold values SA and Sm
Since it exceeds (80, 15), the ANK recognition recognition unit determines and outputs "A", which is the answer of the knowledge result R1, as the final answer. In addition, the six values of recognition results R1 and R2 are the reference data S
If it is smaller than A r SB, the determination circuit 4 determines and outputs "Ryo", which is the answer of the recognition result R2 of the kanji recognition unit 3, as the final answer.

In this way, when recognizing characters in a text consisting of ANK characters and kanji, etc., refer not only to the recognition result R2 of the kanji recognition unit 3, but also to the ANK recognition recognition unit's recognition result R1, which originally has high recognition accuracy for ANK characters. Thus, ANK characters can be recognized with high accuracy. In this case, recognition result R1
Based on this, the determination circuit 4 determines and outputs the final answer using the reference data of the determination table. This determination table is made up of recognition statistical data set in advance so that the optimal answer can be selected based on the similarity value of each recognition result R1, R2 of the ANK recognition recognition unit and the kanji recognition unit 3.

〔Effect of the invention〕

As detailed above, according to the present invention, even when recognizing characters in a text consisting of ANK characters, kanji, etc., in Kanji OCR, the accuracy of recognition of ANK characters is increased to ensure reliable reading. I can do it. Therefore, as a result, it is possible to obtain the effect of increasing the reading accuracy of the entire text recorded on a form or the like.

[BRIEF DESCRIPTION OF THE DRAWINGS] FIG. 1 is a block diagram showing the configuration of a character recognition system according to an embodiment of the present invention, FIG. 2 is a diagram showing an example of a form for explaining the operation of FIG. 1, FIG. 3 is a diagram showing an example of a judgment table for explaining the operation of the judgment circuit shown in FIG. 1. 9- 2...ANK recognition section, 3...Kanji recognition section, 4...
Judgment circuit. Applicant's agent Patent attorney Takehiko Suzue 10-

Claims (1)

[Claims] A first recognition unit that recognizes characters in a limited range of character types such as alphanumeric characters and symbols to be recognized, and a second recognition unit that recognizes characters in a wide range of character types, including the above-mentioned limited character types and kanji, etc. and,
A final answer is determined based on preset recognition statistical data in order to select the optimal answer based on each similarity value from each g recognition result output from the first and second recognition units. A character recognition device characterized by comprising a recognition collation determination unit that outputs an output.