JPH0685190B2 - Dictionary search method for character recognition device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a dictionary search method in a character recognition device.

[Prior art]

In the conventional technology of character recognition, a direction code is given to the contour part of the input pattern, the pattern with this direction code is divided into a plurality of parts, and a histogram is aggregated for each divided part for each direction code. There is a method of determining candidates by summing up the distances of a dictionary (template) for recognition prepared in advance. However, the conventional technique has a problem that it takes time to search the dictionary.

For example, if a pattern with a direction code is divided into 4x4, it becomes 16 blocks, and these are direction code 1 respectively.
Taking histograms for ~ 8 results in 128-dimensional parameters. Distances are calculated for each of these 128-dimensional parameters. At this time, D = Σ | D _H −P _H | <T _H where D is the sum of distances D _H is the histogram parameter of the dictionary P _H is the histogram parameter of the input pattern T _H is the template on which the reject threshold are candidates character, beyond the more times to compare T _H and D, a lot of time taken for the process, also a small, T _H reversed However, since the distance calculation is often performed, the efficiency is low.

〔Purpose〕

An object of the present invention is to shorten the dictionary search time in the character recognition device.

〔Constitution〕

The present invention, the distance reject thresholding the sum D of de T _H
By providing the point to be compared with the character type unit, the character unit or the template unit, the dictionary search time can be shortened. An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the present invention, and particularly shows a block diagram of a portion for performing recognition processing. FIG. 2 is an example of a dictionary configuration. The address of the dictionary 11 is divided into upper and lower addresses. The character code is at address 0 of each lower address, the number of dictionaries (templates) that character code has at address 1, and the character code at address 2 is the leading dictionary (leading template). ) Lower address is stored. Furthermore, D and T _H at 3-5
Stores three points for comparing. The number of check points is generally unlimited.

FIG. 3 shows a processing flow of the microprocessor 10. When performing recognition processing, the microprocessor 10
Sets a dictionary address counter (dictionary counter) 12, fetches the character code, the number of dictionaries, the lower head address, check points 1, 2, and 3 from the dictionary 11, respectively, and stores them in a predetermined register in the microprocessor 10 (step
101). After that, the recognition process is started.

First, the microprocessor 10 is higher than the dictionary counter 12,
Set lower address, checkpoint register
Set the value of checkpoint 1 to 13 (step 10
2). As a result, the first template 1-from the dictionary 11
1 is read. The microprocessor 10 calculates and accumulates the distance between the read template and the parameter of the input pattern (step 103). Next, the value of the dictionary counter 12 is incremented and the same processing is performed (step 104). Thus, the value of the dictionary counter 12 is incremented one after another, and this value and the value of the register 13 are compared with each other by the comparator 14
Compare with. When the value of the dictionary counter 12 matches the value (checkpoint) of the register 13, the comparator 14 issues an interrupt signal to the microprocessor 10.

Upon receiving the interrupt signal from the comparator 14, the microprocessor 10 compares the cumulative distance D with the predetermined reject threshold T _H (step 105). If D ≧ T _H , move to the next dictionary (template) (step 10)
6). If D <T _H , set the next checkpoint 2 in the checkpoint register 13 (step 10
9) Further advance the recognition process. Then dictionary counter 1
When the value of 2 coincides with the checkpoint 2 of the register 13, the microprocessor 10 is again interrupted, and the microprocessor 10 compares the cumulative distance D with the reject threshold T _H. If D <T _H , the checkpoint 3 is set in the checkpoint register 13 and the recognition process is further advanced.

In this way, the candidates that have passed all the check points remain as candidates, and the candidate with the smallest distance is finally selected (steps 107 and 108).

FIG. 4 shows mutually rejected states in a dictionary used for recognition of 7-segment characters. In FIG. 4, a thick underline indicates that many patterns are rejected, and if a checkpoint is placed here, the rejection can be judged efficiently.

In the dictionary configuration shown in FIG. 2, checkpoints are set for each character, but if set for each template, efficiency can be expected to be improved and there is no difference between templates. If so, for example, by setting a checkpoint for each character type, an increase in dictionary capacity can be prevented.

[Effect]

As described above, according to the present invention, the direction code is given to the contour portion of the input pattern, the pattern with the direction code is divided into a plurality of parts, and histograms are aggregated for each divided part for each direction code. , In the character recognition device that determines candidates by summing up the distances of the aggregated histogram and the prepared template for recognition, it is possible to efficiently set the reject points of the dictionary (template) for the input pattern.
You can expect speedup of dictionary search.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a diagram showing an example of a dictionary structure, FIG. 3 is a processing flow chart of the microprocessor of FIG. 1, and FIG. 4 is recognition of 7-segment characters. It is a figure which shows the mutually rejected state in a business dictionary. 10 ... Microprocessor, 11 ... Dictionary, 12 ... Dictionary counter, 13 ... Checkpoint register, 14 ... Comparator.

Claims (4)

[Claims] 1. A direction code is given to the contour portion of an input pattern, the pattern with the direction code is divided into a plurality of parts, and a histogram is collected for each divided part for each direction code. In a character recognition device that determines candidates based on the sum of distances of a prepared dictionary (template) for recognition, a point for comparing the sum of distances with a threshold of a reject is given in advance, and the sum of distances and the reject sled are given for each point. A dictionary search method for a character recognition device, comprising comparing thresholds and determining whether or not to further advance the recognition processing based on the result. 2. The dictionary search method for a character recognition device according to claim 1, wherein the point for comparing the sum of the distances with the threshold of the reject is given for each character type. 3. The dictionary search method for a character recognition device according to claim 1, wherein the point for comparing the sum of the distances with the threshold of the reject is given for each character. 4. The dictionary search method for a character recognition device according to claim 1, wherein points for comparing the sum of the distances with the threshold of the reject are given in units of one template.