JPS5999586A - Optical character reader - Google Patents

Abstract

PURPOSE:To constitute an OCR device which obtains a correct answer by repetitive bulk sorting processing on different condition when the correct answer is not obtained by the 1st bulk sorting processing. CONSTITUTION:A character recognizing circuit 12 decides on a failure of recognition (rejection) and outputs a rejection signal to a selecting circuit 10 when a recognition result is not obtained even by recognition processing by using all pieces of address information stored in the 1st address buffer 8. Namely, the recognition result is not obtained by bulk sorting in this case, but the recognition processing is performed again by using pieces of address information on a feature parameter group whose features are matched resulting from that the number of hits becomes two, namely, that values of two out of three kinds of feature parameters enter a threshold range. The selecting circuit 10 selects address information outputted from the 2nd address buffer 9 by the rejection signal R supplied from the character recognizing circuit 12 and outputs it to the character recognizing circuit 12.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to an optical character reading device using a large classification dictionary method for reading handwritten Chinese characters.

[Technical background of the invention and its problems]

In general, in order to read handwritten kanji with an optical character reader (hereinafter referred to as OCR), the number of characters to be read is approximately 2 compared to the case of alphanumeric characters and kana characters q.
000 to 4000 characters9, requiring a dictionary memory with a large storage capacity. When performing reading processing of handwritten kanji etc. using such OCR, all standard patterns stored in the memory are accessed and processed into kanji to obtain recognition results. This requires a large amount of recognition processing time and is practically impossible.

In order to solve these problems, OCR using a large classification dictionary has been proposed in recent years. This method of OCR is
Various characteristic parameters (for example, area, inflection point, bifurcation point, etc.) are extracted from the input/flat turn of the character to be read. Furthermore, the above-mentioned feature parameters are compared with a dictionary composed of a group of feature parameters for large classification stored in advance in the dictionary memory for large classification, and the large classification process is performed. Through this major classification process, the number of subjects to be processed for recognition of human kataans is significantly reduced, for example, to about 1/10 of the total character types to be read. Then, only the characters to be read in a predetermined range as a result of the major classification process are checked for consistency with the standard pattern in the character recognition dictionary memory in order to obtain a final answer.

However, in the conventional OCR as described above, if there is no correct answer among the narrowed range of characters to be read as a result of broadly classifying the input pattern, the character will be rejected. In this case, it may be possible to obtain an answer by further performing post-processing based on words or sentences, but this would be difficult as the processing content and structure would be extremely complicated.

[Purpose of the invention]

This invention was made in view of the above circumstances, and its purpose is to use a broad classification dictionary method to obtain the correct answer with a simple configuration even if the correct answer cannot be obtained in the first broad classification process. An object of the present invention is to provide an optical character marking device that can improve reading efficiency.

[Summary of the invention]

The present invention provides first and second address buffers that store address information for accessing a dictionary memory for character recognition. The address information corresponding to the feature parameter group is selected and stored in the first and second address buffers based on the hit signal of the feature parameter group based on the major classification process. Furthermore, based on the recognition process, address information is output from either the first or second address buffer in response to a direct signal, and the character recognition dictionary memory is accessed. If the answer is rejected as a result of the recognition process, the address information stored in another address buffer is used to access the character recognition dictionary memory and the correct answer is obtained by performing the recognition process again. .

[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 basic configuration of the OCR according to the present invention, and 1 is a photoelectric conversion circuit that scans a document and converts the recorded characters into a binary signal. The binary signal output from the photoelectric conversion circuit 1 is stored in the line buffer memory 2 as an image for one line on the form. The character cutting circuit 3 detects an image of one character from the contents of the line buffer memory 2 and cuts it out. This cut out image of one character, ie, the character's arm turn, is subjected to major classification processing by collating the various features and parameters of the character one by one in a feature extraction circuit 4. Through this major classification process, the major classification determination circuit 5 receives the hit signal I(
is output to the selection circuit 7. The selection circuit 7 selects this hit (
According to the Q number H, address information outputted from the major classification dictionary memory 6 is selected and outputted to either the first address buffer 8 or the second address buffer 9. The selection circuit 10 selects the address information output from either the first address buffer 8 or the second address buffer 9 depending on the presence or absence of the reject signal R output from the character g function circuit 12, and performs character recognition. output to the dictionary memory 11. The character recognition circuit 12 reads character patterns from the character recognition dictionary memory 11 based on the character pattern cut out from the line buffer output by the major classification determination circuit 5 after the major classification processing and the address information selected by the selection circuit 10. Character recognition is performed by matching with the generated standard pattern, calculating similarity, and outputting a character code as the recognition result. If character recognition is not possible, the character recognition circuit 12 outputs a reject code and sends a reject signal R to the selection circuit 10.
Output to.

In such a configuration, the present invention will be described in detail with further reference to FIG. FIG. 2 shows the main classification dictionary memory 6, first and second address buffers 8, and
9 is a diagram showing in detail the contents stored in the character recognition dictionary memory 1.

Assume now that a character pattern for one character is cut out from the line buffer memory 2 by the character cutout circuit 3 and provided to the feature extraction circuit 4.

The feature extraction circuit 4 obtains feature parameters of character patterns required in the major classification process. This feature parameter is
Normally, about 15 types are required for the four-turn classification process of complex characters such as kanji, but in this example, 3 types are required.
Type.

The major classification determination circuit 5 compares the three types of parameters obtained by the feature extraction circuit 4 with the feature parameters stored in the major classification dictionary memory 6 to perform major classification determination. This major classification dictionary memory 6 stores n feature parameter groups as shown in FIG. That is, n pieces means that they are roughly classified into n types. In FIG. 2, two feature parameter groups are illustrated, namely a feature parameter group 61 and a feature meter group 62.

The feature parameter group is composed of the three types of feature parameters described above and address information. What are the characteristics of this class 38? each parameter has a maximum value and a minimum value,
This constitutes the threshold value. The address information is the address of the character recognition dictionary memory 11 in which standard patterns of characters roughly classified by the three types of feature parameters constituting the feature i + parameter group are stored, and the number is the feature parameter. It varies depending on the group.

The major classification determination circuit 5 compares whether the three types of feature parameters obtained by the feature extraction circuit 4 are within the threshold value of the corresponding feature A' parameter of each feature parameter group in the major classification dictionary memory 6.

Now, let us consider a comparison between the feature parameters obtained by the feature extraction circuit 4 and the feature parameter group 61. First, the obtained first feature parameter and the corresponding feature or parameter t are compared.

Subsequently, the second feature parameter is compared with the feature meter m, and similarly the third parameter is compared with the feature parameter n. As a result of this comparison, if the value of the feature parameter found is within the range of the threshold value, it is said to be a hit. The major classification determination circuit 5 outputs a hit signal H corresponding to the number of hits to the selection circuit 7. At the same time, the address information of the compared feature parameter group is output to the selection circuit 7. In the selection circuit 7, when the number of hits is 3, the address information output from the major classification dictionary memory 6 is output to the first address buffer 8 using the hit signal H output from the major classification determination circuit 5. , store it there. If the number of hits is two, the selection circuit 7 outputs the address buffer to the second address buffer 9 and stores it there. Further, when the number of hits/number is 1 or O, the selection circuit 7 closes the dart and does not output the given address information. Thereafter, in a similar manner, the major classification determination circuit 5 compares all feature parameter groups stored in the major classification dictionary memory 6, and depending on the result, the address information is transferred to the first address buffer 8 or the second address buffer. The data is stored in one of the address buffers 9 of. When all comparisons are completed, the character pattern given from the feature extraction circuit 4 is output to the character recognition circuit 12.

Well, all comparisons have been completed and the 1st and 8i! The following operation will be explained assuming that address information is stored in the address buffers 8 and 9 of No. 2 as shown in FIG. Currently, the eject signal R is not output from the character recognition circuit 12, so the selection circuit 10 transfers the address information output from the first address buffer 8 to the character recognition dictionary memory 11.
given to. The character recognition circuit 12 first reads out a standard pattern and its corresponding character code from the character recognition dictionary memory 11 based on the address information e. The character recognition circuit 12 performs matching between the read standard pattern and the character pattern cut out by the character cutout circuit 3 to calculate the degree of similarity. As a result of this calculation, if the similarity value exceeds a predetermined value, the character code stored in association with the standard e-turn is outputted to the outside as a recognition result. If the similarity value does not exceed the P)r fixed value, the next address information is read out from the first address buffer 8 and the same recognition process is performed. When the character recognition circuit 12 cannot obtain a recognition result even after performing recognition processing using all of the address information stored in the first address buffer 8, the character recognition circuit 12 outputs a reject signal to the selection circuit 10 as unrecognizable (reject). do.

Rejected means that no recognition result can be obtained by major classification.

Therefore, next time, recognition processing is performed using the address information of the feature parameter group in which the number of hits is 2, that is, the values of 2 of the 3 types of % characteristic parameters are within the range of the % characteristic parameter, and the features match. do. The selection circuit IQ selects the address information output from the second address buffer 9 and outputs it to the character recognition circuit 12 in response to the eject signal R applied from the character recognition circuit 12 . Thereafter, in the same way as described above, the character recognition circuit 12 sequentially performs recognition processing using the standard pattern read from the character recognition dictionary memory 11 with the address information stored in the second address buffer 9. If the second
If a recognition result cannot be obtained using all the address information stored in the address buffer 9, a reject code is output as an unreadable character. In this embodiment, a similarity method using a standard pattern is used as a character recognition method, but other methods may be used.

In this way, even if the final answer cannot be determined in the first major classification process, the character recognition dictionary memory 11 is accessed using the address information of the feature parameter group with fewer feature parameters matching the first major classification. . The character recognition circuit 12 performs the recognition process again and outputs the recognition result.

Therefore, if the correct answer cannot be determined in the first major classification process, the rate of determining the correct answer can be increased by performing the second major classification process with a relaxed condition on the number of hits. Therefore, character reading efficiency can be improved without performing complicated reading processing such as word and sentence processing. −
[Effects of the Invention] As detailed above, according to the present invention, major classification terms 1. In this type of OCR, even if a correct answer is not obtained in the first major classification process, the rate of obtaining a correct answer can be increased by performing the major classification process again under different conditions. Therefore, it is possible to improve the reading efficiency of characters such as Chinese characters with a simple configuration without adding particularly complicated circuits and processing.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an optical character reading device according to an embodiment of the present invention, and FIG. 2 shows the main classification dictionary memory 6 and the first and second address buffers 8 in FIG.
, 9 is a diagram showing in detail the contents stored in the character recognition dictionary memory 1. 1... Photoelectric conversion circuit, 2... Line buffer memory, 3... Character cutting circuit, 4... Feature extraction circuit,
5... Major classification determination circuit, 6... Major classification dictionary memory, 7... Selection circuit, 8... First address buffer, 9... Second address buffer, 10... Selection circuit, 11... Dictionary memory for character recognition, 12... Character recognition circuit.

Claims (1)

[Claims] In an optical optical character reading device using a broad classification dictionary method, there is a large classification dictionary memory that stores in advance each characteristic number and parameter group for large classification processing, and large classification is performed based on each feature parameter group of this large classification dictionary memory. Based on the comparison result between the first and second address buffers that store address information for accessing the character recognition dictionary and memory after processing, and the feature parameter group based on the above-mentioned major classification, the above-mentioned selection means for returning and storing the address f+' corresponding to each characteristic parameter group in either one of the first and second address buffers; An optical character reading device comprising: address output means for selecting output of address information from either one of the second address buffers.