JPH03212783A - Matching system - Google Patents

Abstract

PURPOSE:To maximize a processing time shortening effect based upon a dymanic threshold and to speed up matching processing by executing matching processing in the order of classes determined in accordance with the recongized character sort of the character preceding the current one to be recognized. CONSTITUTION:A control circuit 10 determines the class order for executing the matching processing in accordance with the recognized character sort of the character preceding the current one to be recognized, writes respective class numbers and the number of templates in an FIFO memory 11 in a dictionary address forming circuit 3 and then writes the feature pattern of the current character image to be recognized in a feature pattern memory 1. One- dimensional data in the feature pattern are read out from the memory 1, the data of one dimension corresponding to one template belonging to one class are read out from a dictionary memory 2 and respective read data are set up in a distance table address register 5. A distance table memory 4 outputs distance data based upon the value of the register 5 and the circuit 10 allows a sorting circuit 8 to sort distance data and the character code outputted from the memory 2 by using the distance data as a key to update candidate data.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention identifies candidate characters with a small distance by matching the feature patterns of character images to be recognized with standard feature patterns (templates) stored in a dictionary. The present invention relates to a character recognition device using a method (dynamic threshold method) in which the distance of the lowest candidate character at that time is used as a threshold value to decide whether to terminate the distance calculation of the matching process.

[Prior art and problems to be solved by the invention] This type of character recognition device can reduce unnecessary distance calculation time and speed up matching processing compared to a method that fixes the abort judgment threshold for distance calculation. . However, even with the dynamic threshold method, in Kanji OCR, etc., where the number of characters and character features are extremely large,
The reality is that Lg recognition processing speed is insufficient.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a matching method that maximizes the time-saving effect of the dynamic threshing method and achieves faster character recognition processing.

As a means to achieve this objective, we collect characters with similar feature patterns and classify them into classes, prepare class feature patterns that represent the feature patterns of characters belonging to each class, and first, we create class feature patterns that represent the feature patterns of characters belonging to each class. The classes are prioritized in descending order of distance and distance by matching with class feature patterns (a kind of major classification), and the standard feature patterns (templates) of the characters belonging to the class are determined in descending order of distance and distance. A method for performing matching has been devised (Japanese Patent Application No. 63-275133).

According to this method, candidate characters with a high probability of being correct are obtained early, the threshold value for determining whether to discontinue distance calculations is quickly reduced, and an improvement in the effect of saving wasted distance calculation time can be expected.

However, on the other hand, a memory for the class feature pattern is required, and the matching process with the class feature pattern and the matching process with the standard feature pattern of characters for each class have to be handled separately, making the control complex. is unavoidable.

Another object of the present invention is to provide a matching method that maximizes the effects of the dynamic threshing method while eliminating such complication of control and increase in memory capacity.

[Means for Solving the Problems] The matching method of the present invention classifies a dictionary by character type in a dynamic threshold character recognition device, and performs matching processing on each recognition target character image based on the recognition result of the previous character. It is characterized by being executed in the order of classes determined by the character type.

[For production]

In Japanese documents, there is generally a high possibility that the same character type will appear consecutively. Therefore, according to the matching method of the present invention, for example, candidate characters with a high probability of being correct in most cases are selected by performing matching processing with priority given to the class corresponding to the character type of the upper candidate character of the immediately preceding recognized character. Since the distance calculation can be obtained at an early point in time and the threshold value for discontinuing the distance calculation can be made sufficiently small, the wasted distance calculation time for characters with a small probability of being correct is greatly reduced, and the matching processing time is also significantly shortened.

In addition, since the matching process for major classification is not performed separately from the detailed matching process, there is no need to worry about complicating the control.
Furthermore, no storage memory for class feature patterns is required.

〔Example〕

FIG. 1 is a block diagram of a matching processing system according to an embodiment of the present invention. 1 is a feature pattern memory in which feature patterns obtained by extracting features of the recognition target image are stored; 2 is a standard feature pattern (template) for each character;
3 is a memory address generation circuit that generates addresses for the dictionary memory 2 and feature pattern memory 1. 4 is a distance table memory for obtaining distance data for each dimension of the feature pattern and template; 5 is a distance table address register which synthesizes the corresponding one-dimensional data of the feature pattern and template and provides it as an address to the distance table memory 4; be. 6 and 7 are adders and accumulators for cumulatively adding distance data for each dimension.

8 is a sorting circuit that sorts and holds the character code and distance data in ascending order of distance value; 9 is a distance calculation discontinuation judgment based on dynamic thresholding, that is, the distance data obtained in the accumulator 7 and the lowest candidate obtained in the sorting circuit 8; This is a comparison circuit that performs comparison judgment with character distance data. Reference numeral 10 denotes a control circuit that controls the entire matching processing system, communicates with the outside of the system, and determines the class order in which matching processing is executed based on the candidate character type of the immediately preceding character. The priority order of the classes is determined, for example, so that the class corresponding to the character type of the candidate character higher than the immediately preceding character is given priority. However, it is also possible to adopt a method in which the class corresponding to the character type of the highest candidate is ranked first, and the ranks of the remaining classes are fixed.

FIG. 2 is a block diagram of the memory address generation circuit 3. 11 is a FIFO memory that stores class numbers and the number of templates belonging to them in the order of matching processing; 12
13 is a class number register that indicates the position of the class in the dictionary that is being matched, and 13 is a template number counter that indicates the position of the template that is being matched within the class. , 14 is the lowest address counter indicating the number of data (dimension number) in the template, and 15 is an in-class template counter that holds the number of remaining templates in the class during which matching processing is being executed.

In addition, within the control circuit 10, there is a dimension counter I that holds the number of remaining bytes (dimensions) of the feature pattern to be processed.
There is OA.

Next, the operation will be explained.

Operation steps ■ The control circuit 10 determines the classes to be matched according to the character type of the recognition result of the previous character, and in accordance with that order, transmits the number of each class and the number of templates to the dot IF in the dictionary address generation circuit 3. ○Write into the memory 11, and then write into the feature pattern memory 1 the feature pattern of the character image to be recognized this time.

Operation Steps ■ The control circuit 10 initializes the template number counter 15 and the lowest address counter 14 in the dictionary address generation circuit 3, reads out the number of the class to be processed and the number of templates from the FIFO memory 11, and stores them in the class number register. 12 and the in-class template counter 15.

The feature pattern memory 1 is addressed by the value of the lowest address counter 14, and the dictionary memory 2 is addressed by the composite value of the class number register 12, template number counter 13 and lowest addressing counter 14.

Operation step ■ One dimension of the feature pattern (from feature pattern memory 1)
The data of one template (specified by the template number counter 13) belonging to one class (specified by the value of the class number register 12) is read from the dictionary memory. The data of one corresponding dimension (specified by the value of the lowest address counter 14) of the data table (specified by the value of the lowest address counter 14) is read and set in the distance table address register 5, respectively.

Distance tape memory 4 is distance table address register 5
Distance data is output by being addressed by the value of . This distance data is added to the value of the accumulator 7 by the adder circuit 6, and the result is held in the accumulator 7, and is compared with the distance data of the lowest candidate character by the comparison circuit 9.

The accumulator tuff value is the distance data of the lowest candidate character (
If the value is smaller than the determination threshold (determination threshold), the control circuit 10 increases the value of the lowest address counter 14 by 1, decreases the value of the dimension counter IOA by 1, and returns to operation step (3) to start processing of the next dimension. When the dimension counter IOA becomes 0, the control circuit 10 (
When the distance calculation for one template is completed up to the final dimension, the control transitions to operation step ■.
When the value of the accumulator 7 exceeds the determination threshold (when it is determined that the process is to be terminated midway), the control transitions to operation step (2).

Using the distance data in the accumulator 7 as a key, the control circuit 10 causes the sorting circuit 8 to sort the distance data and the character codes output from the dictionary memory 2, thereby updating the candidate data.

Operation steps■ The control circuit 10 is the lowest address counter 14.
Then, the dimension counter IOA is initialized, the value of the template number counter 13 is incremented by 1, the value of the intra-class template counter 15 is decremented by 1, and the control shifts to operation step (2). However, when the value of the intra-class template counter 15 becomes 0, control is transferred to operation step (2) and matching processing for the next class is started. If there is no class left to process next, the process ends.

〔Effect of the invention〕

As explained above, according to the present invention, focusing on the character of Japanese documents in which the probability that the same character type continues is high, the dictionary is divided into classes according to five character types, and the classes are classified according to the character type of the recognition result of the previous character. By executing the matching process, it is possible to reduce the distance calculation abort judgment threshold early, maximize the processing time reduction effect of the dynamic sledge unit, and greatly speed up the matching process. This solves the problems of complicating control and increasing memory, which were problems associated with matching methods.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a matching processing system according to an embodiment of the present invention, and FIG. 2 is a block diagram of an address generation circuit. 1... Feature pattern memory, 2... Dictionary memory. 3...Memory address generation circuit, 4...Distance table memory, 5...Distance table address register, 6...Adder, 7...Accumulator, 8...Sorting circuit, 9... Comparison circuit, 10... Control circuit, 10A... Dimension counter, 11... FIFO memory, 12... Class number register, 13... Template number counter, 14... Lowest address counter, 15 ...Intra-class template counter.

Claims (1)

[Claims] (1) Character recognition that determines candidate characters by matching the feature pattern of the character image to be recognized with the standard feature pattern registered in the dictionary, and aborts the distance calculation in the matching process using the distance of the lowest candidate character as the determination threshold. A matching method characterized in that, in the device, a dictionary is divided into classes according to character types, and matching processing for each character image to be recognized is executed in the order of classes determined by the character type of the recognition result of the previous character.