JPH0259884A - Character recognizing device - Google Patents

Abstract

PURPOSE:To shorten a processing time without decreasing the amount of data on standard characters which effect a recognition rate and to speed up character recognition by calculating the distance to a standard character when the distance to a character being calculated becomes larger than the distance to the least significant digit candidate character. CONSTITUTION:A comparing means 300 compares the distance to the character being calculated by a distance arithmetic means 100 with the distance to the least significant digit candidate character in a sorting means 200 and calculates the distance to the next standard character when the distance to the character being calculated becomes larger than the distance to the least significant digit candidate character. Namely, when the distance being calculated by the distance arithmetic means 100 becomes longer than the distance to the least significant digit candidate character in the sorting means 200, the probability that the standard pattern being calculated is a correct character is equal to zero. Consequently, subsequent calculation is quitted at this point of time and the distance to the next standard character is calculated to shorten the processing time without decreasing the amount of data of of the standard characters, thereby speeding up the character recognition.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a character recognition device that determines candidate characters by comparing input characters with standard characters stored in a dictionary in advance.

In recent years, with the development of office automation, character recognition has been widely studied and some of it has been put into practical use. In character recognition, it is desired to obtain a high recognition rate in a short recognition time.

[Conventional technology]

FIG. 3 is a block diagram of a conventional character recognition device.

A feature amount memory 1 stores feature amounts of characters to be recognized (input characters). Normally, one character is usually divided into 64 to 256 parts,
For example, contour tracking, direction vector assignment, and histogram creation are performed for each divided region to extract feature amounts. The dictionary memory 2 stores the special character tIim of each character (standard character) that becomes a dictionary. The standard character is similarly divided into eight parts, 64 to 256, and the feature amount is determined for each divided area. The address generation circuit 3 sequentially generates and supplies addresses to the feature memory 1 and the dictionary memory 2.

The distance calculation circuit 4 compares the feature 8 read from the feature memory 1 with the feature amount read from the dictionary memory 2, and calculates the distance between them. The control circuit 5 controls the address generation circuit 3, the distance calculation circuit 4, and the sorting circuit 6.

The sorting circuit 6 sorts the candidate characters of the distance calculation model in ascending order of distance or in descending order of distance. The sorting memory 7 is a work memory used when the sorting circuit 6 performs sorting.

Further, the buses 8 and 9 send the feature amounts read from the feature amount memory 1 and the dictionary memory 2, respectively, to the distance calculation circuit 4. The bus 10 sends the distance calculation result and character code for one character to the sorting circuit 6. The bus 11 exchanges distance calculation results and character codes between the sorting circuit 6 and the sorting memory 7 during sorting. Address lines 12 and 13 send addresses to the feature memory 1 and the dictionary chain 2, respectively. Control lines 14, 15 and 16 send control signals to the address generation circuit 3, distance calculation circuit 4 and sorting circuit 6. Address line 17 supplies addresses to sorting memory 7.

Processing for one dictionary character is performed by adding 32 iFi to the distance between the feature quantities read from the special@hanging memory 1 and the dictionary memory 2 for every 64 to 256 divisions, and sequentially adding them. Then, when the addition results of the feature amounts of all the divided regions are obtained, the distance calculation circuit 4 sends the distance calculation results and their character codes to the sorting memory 6, and then performs the distance calculation with the next standard character. Move. The sorting circuit 6 can store a predetermined number of candidate characters, and compares the distance calculation result of the character code just received with the distance of the current candidate character, and uses the sorting memory 7 to sort them in descending order of distance. Or sort in descending order. At this time, the sorting circuit 6 deletes the candidate character with the greatest distance.

The candidate character with the greatest distance after deletion becomes the lowest candidate character.

FIG. 4(A) is a diagram showing the time relationship between distance calculation processing and sorting processing. Di indicates distance calculation processing for one character, and Si indicates one sorting processing. When the distance calculation process for the first standard character is completed,
Next distance calculation process D2 and first sorting process S
+ starts゛. When this is completed, the next distance calculation process D3 and the second sorting process S2 start. Similarly, while the distance calculation circuit 4 is performing the distance calculation WDi for the i-th standard character, the parallel processing of performing the (i-1>th sorting process S1) is performed for all (n) This is performed on standard characters, and the top candidate character with the shortest distance is finally obtained as the recognition result.

[Problem to be solved by the invention]

However, in the above conventional technology, the characters to be recognized are
To recognize a character, distance calculations are performed by comparing all the features of n standard characters with the corresponding recognition target character, so the problem is that the amount of data to be compared becomes enormous and the processing time takes a long time. there were.

If this problem is attempted to be solved by reducing the amount of data such as the number of features of standard characters or the number of standard characters, another problem arises in that the recognition rate deteriorates.

Therefore, an object of the present invention is to solve the above-mentioned problems and speed up character recognition by shortening processing time without reducing the amount of standard character data that affects the recognition rate.

[Means to solve problems]

FIG. 1 is a block diagram of the principle of the present invention.

The distance calculating means 100 calculates the distance between the feature amount of the input character and the characteristic coloring of the standard character. The sorting means 200 sorts candidate characters in descending order of obtained distance or in descending order of distance. The comparison means 300 is the distance calculation means 10
0, the distance of the character being calculated is compared with the distance of the lowest candidate character in the sorting means 200, and when the distance of the character being calculated becomes greater than the distance of the lowest candidate character, the distance between the character and the next standard character is compared. Perform distance calculations.

[Effect]

When the distance being calculated by the distance calculation circuit 100 becomes greater than the distance of the lowest candidate character in the sorting means 200, the probability that the standard pattern being calculated is the correct character is equal to zero.

Therefore, by stopping the calculation that continues at this point and moving on to calculating the distance to the next standard character, the data load for the standard character will not be reduced (in other words, the recognition rate will not deteriorate).
, it is possible to shorten processing time and speed up character recognition.

(Example) Hereinafter, an example of the present invention will be described in detail with reference to the drawings.

FIG. 2 is a block diagram of an embodiment of the present invention. The feature amount memory 21 and the dictionary memory 22 are respectively the same as the feature area memory 1 and the dictionary memory 2 in FIG. The distance calculation circuit 24 is the same as the distance # calculation circuit 4 in FIG. In addition to the functions of the sorting circuit 6 shown in FIG. 3, the sorting circuit 26 has a register (not shown) for storing the distance of the lowest candidate character. Sorting memory 27 is the same as sorting memory 7 shown in FIG. The control circuit 25 controls the address generation circuit 23, the distance calculation circuit 24, and the sorting circuit 26, based on instructions from the comparison circuit 28, as will be described later. The comparison circuit 28 is newly provided in this embodiment, and the distance calculation circuit 24 calculates the distance between the feature amounts for each divided area and sequentially adds the distance, and the sorting circuit 2
6, and the comparison result (carry signal) is sent to the control circuit 25.
Output to.

Address lines 33 and 34 are identical to addresses 112 and 13 in FIG. 3, respectively. The bus 31 sends the distance calculation results to the sorting circuit 26 and the comparison circuit 28.

The bus 32 is used for exchanging data between the sorting circuit 26 and the sorting memo 92, and for comparing the lowest candidates stored in the registers of the sorting circuit 26 with the comparing circuit 28.
send to Control lines 35, 36 and 37 transmit corresponding control signals to the address generation circuit 23 and distance calculation circuit 24, respectively. The address signal 38 is the sorting memory 2
7, the address is transmitted at the time of sorting. The signal line 39 transmits to the control circuit 25 a carry signal generated when the distance being calculated from the distance calculation circuit 24 becomes greater than the distance of the lowest candidate character.

Next, the operation of this embodiment will be explained.

For each divided area, feature quantities are read from the feature quantity menu 21 and the dictionary memory 22, and the distance calculation circuit calculates the distance between the two, and the past distance calculation result of the standard character currently being processed is calculated. will be added to. The distance calculation results thus obtained up to the present time are sent to the comparison circuit 28. The comparator circuit 28 uses this distance calculation result up to the present time and the current most recent one read from the register in the sorting circuit 26.
Compare the distance between the position candidate characters. When the distance calculation result up to now is smaller than the distance of the lowest candidate character, the comparison circuit 28 does not output a carry signal. Therefore, the control circuit 25 gives an instruction to the address generation circuit 23 to update the address. Thereby, the distance between the feature floors of the next divided area is calculated.

On the other hand, when the distance calculation result up to now is greater than the distance of the lowest candidate character, the comparison circuit 28
A carry signal is output for 5. The control circuit 25 is
This carry signal instructs to jump to the address where the next standard character is stored in the dictionary memory 22, and also transfers the address given to the feature memory 21 to the feature of the first divided area (first the distance Give an instruction to set it to the address where the characteristic seedling to be calculated) is stored. As a result, as soon as it is determined that the standard character currently being processed is not a candidate character, processing of the next standard character is started.

The state of this processing is shown in FIG. 4(B). In the figure, Qi, Q
i+2 and D1+3 indicate distance calculation processing for standard characters that are found to be excluded from candidate characters during distance calculation. In this case, the sorting process that is performed in parallel is not executed, and a waiting state occurs. For example, Di after DI
Sorting process S1,011 for distance calculation process of +1
passes through a waiting state and is started simultaneously with the execution of the distance calculation process of Di+2.

In this way, according to the present embodiment, it is not necessary to compare all the feature amounts of the n standard characters for distance calculation, so it is possible to speed up the processing without deteriorating the recognition rate.

〔Effect of the invention〕

As explained above, according to the present invention, when calculating the distance between an input character and a standard character, from the time when it is determined that this standard character cannot be a candidate character for the input character, the subsequent distance of this standard character is calculated! ! Since the distance 1111f calculation for the next standard character is immediately performed without executing 1tJf'f, the processing time can be shortened without reducing the data size of the standard character, or in other words, without deteriorating the recognition rate. Character recognition can be speeded up.

The effect of the present invention becomes extremely significant as the number of standard characters in the dictionary memory exceeds the predetermined number of candidate characters.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a block diagram of an embodiment of the present invention, Fig. 3 is a 10-step diagram of a conventional character recognition device, and Fig. 4 (A) is a conventional distance calculation method. FIG. 4B is a diagram for explaining the relationship between the processing and the sorting process, and FIG. 4B is a diagram for explaining the relationship between the distance calculation process and the sorting process according to the embodiment of the present invention. 24 is a distance calculation circuit, 25 is a control circuit, 26 is a sorting circuit, 27 is a sorting memory, and 28 is a comparison circuit. Patent Applicant Fujitsu Ltd. In the figure, 21 is a feature memory, 22 is a dictionary memory, 23 is an address generation circuit, Figure 1. Figure 2 Figure 3

Claims (1)

[Scope of Claims] Distance calculating means (100) for calculating the distance between the feature amount of an input character and the feature amount of a standard character; and a sorting means for rearranging candidate characters in order of decreasing or increasing distance obtained. (200), the distance of the character being calculated by the distance calculation means (100) and the distance of the lowest candidate character in the sorting means (200) are compared, and the distance of the character being calculated is the distance of the lowest candidate character. A character recognition device comprising: a comparison means (300) that controls a distance calculation means (100) to calculate the distance to the next standard character when the character becomes larger.