JPH0640352B2 - Character recognition device - Google Patents

Description

Description: TECHNICAL FIELD OF THE INVENTION The present invention relates to a character recognition device or a voice recognition device capable of easily and effectively recognizing, for example, a handwritten or voice input word consisting of n characters. It is about.

[Technical Background of the Invention and Problems Thereof] When recognizing a character, basically, the characteristic of the character is detected, but when recognizing a word consisting of n characters,
It is often not enough to recognize each individual character and combine the results, and if the recognition of a certain character is insufficient, it becomes difficult to recognize the word. Therefore, conventionally, even when individual character recognition is insufficient, a word collating unit is configured as shown in FIG. 4 and a word dictionary in which significant words are stored is searched to search for that word. It has been devised to recognize it (Japanese Patent Application Sho 56).
-See 138163). In this case, n characters forming a word are first recognized by the character recognition unit I, and the result B _j , i (i = 1, n; j = 1, k) is stored in the candidate character register 5. . Here, B _j , i is the character code of the j-th candidate character when the character at the character position i in the word is recognized. Next, when the comparator 6 compares the character code of the word stored in the word dictionary memory 4 with the character code, and when these match outputs are obtained, the degree of matching is determined based on the candidate rank. The calculation unit 9 calculates the degree of coincidence between this word and the input word.

However, in this method, the candidate character register 5 must be searched for the number of candidates k times for each character in order to carry out a character code matching test, which requires time and processing time, and the control circuit It was complicated and had problems.

In order to solve this drawback, the candidate character register 5 and the comparator 6 shown in FIG. 4 are conventionally used (see Japanese Patent Application No. 58-071372).
Instead, the circuit shown in FIG. 5 is used to calculate the degree of coincidence. In this method, the address of the similarity memory 17 is directly referred to by the character code of the word data 13, and the output of this memory is sent to the matching degree calculation unit 18 to perform the matching degree calculation. In the similarity memory 17, the similarity obtained from the recognition result of the character corresponding to each character code is stored at the address indicated by this character code. However, with this method, if a character code having a large number of bits, for example, a kanji code (16 bits) is to be handled, the similarity memory as a word becomes huge and it becomes impossible to realize.

[Object of the Invention]

This invention was made in consideration of such circumstances.
The purpose is to enable efficient word recognition by easily and quickly comparing the word dictionary with the input word even when handling a character code having a large number of bits such as a kanji code. To provide a character recognition device.

[Outline of Invention]

This invention uses each character A _i (i =
1, 2, ..., N) are respectively recognized, and the character A _i for each character and the recognition target character B _k (k = 1, 2, ... L) registered in L characters in the character recognition dictionary The similarity S _k , i is obtained, and this S _k , i is stored in the second memory which is a two-dimensional memory. Then, when the character code of the character B _k is divided and input into the first memory divided into a plurality of pieces, an address pointer indicating the storage location of this S _n , i is written, and at the time of word recognition, a word dictionary Output character code and character position i
Is divided into the addresses of the first memory, the resulting address and the character position i are input to the second memory, and the output is sent to the matching degree calculation unit to calculate the matching degree of the words.
A word that is a candidate for recognition is carefully selected on the basis of the degree of coincidence, thereby enabling a means for easily performing accurate recognition at high speed.

〔The invention's effect〕

As described above, according to the present invention, the similarity search in the coincidence calculation is performed only by serially referring to the first memory and the second memory, so that the control circuit is simplified and each memory is Since the address is divided input, it can be made small, and the recognition processing time can be greatly shortened, which has the effect of enabling extremely effective and highly practical character recognition.

Example of Invention

An embodiment of the present invention will be described below with reference to the drawings. In the following description, the number of divisions of the first memory will be two.

FIG. 2 shows a character code string obtained as a result of character recognition.
The position candidate is a character candidate of M ₁ N ₁ , and the jth position is M _j N _j . Here, M ₁ is the upper 8-bit code of the character code, N
₁ is the lower 8-bit code. That is, in this example, the character code M _j N _j is 16 bits, and each division is 8 bits. Next, these M _{1 to} M ₁₅ (the maximum number of candidates is limited to 15) are assigned numbers _{1 to 15} , and if the same code exists in M _{1 to} M ₁₅ at this time, Assign the same number. That is, M _j = M _k (j
If <k), the number J is assigned to both M _j and M _k . This number
Described as JM _j . The same applies to N _j . Next, the above-mentioned corresponding number JM _j is written in the position of M _j (8 bits) of the first memory [M] (FIG. 22) as an address. N _j of the other first memory [M] (FIG. 3)
Similarly, the corresponding number JN _j is written in a position having (8 bits) as an address. The rest of the first memory is all 0s
Is. Further, the similarity S _j corresponding to the jth candidate M _j N _j is set to M _j
Storing the corresponding number JM _j, address JM _j produced by a set of numbers JN _j corresponding to N _j, the address of the second memory as defined in JN _j to. That is, the input characters A _i are actually B
_The similarity S _j that is _ji is the two-dimensional matrix S of (JM _j , JN _j ).
_When it is displayed on _{JM and JN} , it is written like the second memory 28 in FIG. The rest of the second memory 28 is all zeros. Incidentally, 26 and 27 in FIG. 3 are the second memory 28.
The upper 4 bits and the lower 4 bits of the address are displayed. By doing so, when the character codes of M _j N _j are input to the respective addresses of the first memories [M] and [N] through the signal lines 20 and 21, the first memories are JM _j and JN. outputs _j becomes numbers each signal line 24, which is input to the address of the second memory, the similarity S _j corresponding from the second memory to M _j N _j is obtained. on the other hand,
When a code other than M _j N _j (j = 1, 2, ... 15) is input, the second memory output becomes 0.

The character position i is input to the first and second additional addresses m by adding m bits to the addresses of the first and second memories, respectively. By doing so, the similarity S _j is obtained by the above-described processing for each character i of the word consisting of n characters.

The degree of similarity of each character thus obtained is sent to the degree-of-coincidence calculator 29, and, for example, the degree of similarity, that is, degree of coincidence, of the word is obtained by adding these degrees of similarity. This degree of matching is calculated between one word registered in the dictionary and the input character, and this is calculated for all the words in the dictionary. For example, the candidate word with the highest degree of matching is output to the outside as a result. Output.

FIG. 1 is a schematic configuration diagram of an embodiment of the present invention, and its operation is as follows. That is, the result of the character recognition unit 31 is the result of the above-described processing in the first and second memories 33 and 3.
The character code written in 6 and output from the word dictionary memory 32 is cascaded to the first memory 33, the second memory 36 (the same as the second memory 28 in FIG. 3), and the coincidence calculation unit 38. Is sent to and the degree of coincidence is calculated. The control unit has means for controlling the address counter 34 and the character position counter 35 and also for controlling the whole, and sending a result based on the similarity to the character recognition control unit 30.

The present invention can be implemented in various modifications other than the above-described embodiments. For example, the number of divisions of the first memory is arbitrary, and the number of bits of the address is also arbitrary. The method of calculating the degree of coincidence is also arbitrary, and any device and means can be applied to the present invention. Also, the maximum number of candidates at the time of character recognition is not particularly limited. Also, the means to be realized are not limited to specific hardware.

It should be noted that what is described as characters in the above description includes both a handwritten input pattern and a voice input pattern.

Further, the similarity on a character-by-character basis may be any value as long as it indicates the likelihood of a character, and may be, for example, a candidate rank at the time of character recognition, or a score based on the rank.

Further, a system for directly sending the output of the first memory to the coincidence degree calculating unit may be added to the present invention so that a direct system is possible when the bit number of the character code is small.

In short, the present invention can be variously modified and implemented without departing from the gist thereof.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of the present invention, FIGS. 2 and 3 are diagrams for explaining each means of the present invention, and FIGS. 4 and 5 are schematic configuration diagrams showing an example of a conventional device. Is. In the figure, 1,31 ... Character recognition unit, 2 ... Editing unit, 3,39 ...
Word collating unit, 4, 12, 19, 32 ... Word dictionary memory, 5 ... Candidate character register, 6 ... Comparator, 7
...... Word dictionary register, 8 ...... Register counter, 9,
18, 29, 38 ... Matching degree calculation section, 10, 34 ... Address counter, 11 ... Sort processing section, 13 ... Word data, 14, 15, 16 ... Similarity storage position, 17 ...
... Similarity memory, 20, 21, 24, 25 ... Signal line,
22, 23, 33 ... First memory, 28, 36 ... Second memory, 26 ... Lower address of second memory, 2
7 ... Upper address of second memory, 30 ... Character recognition control unit, 35 ... Character position counter, 37 ... Control unit.

Claims (1)

[Claims] 1. A character recognition means for performing character recognition on a voice input word, which is given as a character string consisting of n characters, or each character of the character input word, and a similarity obtained at the time of processing by the character recognition means. Second memory means for storing, a plurality of first memory means for dividing and storing an address referencing the second memory means, and a word based on the similarity of each character stored in the second memory means. And a means for selecting a word as a recognition candidate from the word dictionary based on the result obtained by the coincidence degree calculating means. The first memory means has a plurality of character codes of the respective characters of the word registered in the word dictionary as addresses, and stores the addresses referencing the second memory means in a divided manner. dictionary The selection of words as recognition candidates is performed by referring to the plurality of first memories for each character of the word in the word dictionary, synthesizing each output, and using the synthesized output as an address in the second memory. The character recognition device is characterized in that it is performed based on the matching degree of words calculated using the similarity of each character.