JPS5847066B2 - character recognition device - Google Patents

Description

[Detailed Description of the Invention] The present invention compares the characteristic bit pattern of handwritten characters with the contents of a dictionary and outputs a plurality of answer candidates. On the other hand, by creating a character table to be read and comparing it with this, the final result is obtained. This relates to a character recognition device that obtains answers.

Conventionally, this type of character recognition device, as shown in FIG. 1, uses optical character reading (
The data is read by the scanner of the reading mechanism 2 of the OCR) device 10 and stored in the memory 3 as video data.

The stored video data is sent to the feature extractor 4 and immediately analyzed, the features of the character are extracted, and a bit pattern of the read character is created and stored in the register 5.

On the other hand, in the memory in the OCR device 10, the characteristics of many handwritten numbers, kana, alphabets, and various symbols are extracted in advance and stored in the form of bit patterns.

This stored bit pattern group is called a dictionary 6.

Immediately after the bit patterns of the handwritten characters read from the form 1 are created, point-by-point comparison with the bit patterns of the characters in the dictionary 6 is started.

If the bit pattern of the character read here matches the bit pattern of a certain character in the dictionary, that character in the dictionary 6 is output as the reading result.

Further, if the bit pattern does not match any character in the dictionary 6, the character is output as a reject character.

Many of the current character recognition devices have numbers, kana, alphabetic characters, special symbols, etc., either singly or in combination with some of them, but in the latter case, there are many types and combinations of similar characters.

For example, the number "7", the kana "su", "wa", the number "8", and the alphabetic letter rBl are often difficult to distinguish when written by hand.

However, in the conventional feature extraction unit, a determination is made only by comparison with features stored in a dictionary, and a single answer is output.

In this case, even if the character to be read is clearly known to be a numeric character, if there is only one dictionary, the character can be determined only from the shape of the character, so similar characters such as kana may be output.

In order to prevent such errors, correct characters can be recognized by preparing a plurality of dictionaries depending on the object to be read and selecting the dictionary to be read in advance.

However, since a dictionary is a memory with a large capacity consisting of character bit patterns, it is obvious that controlling a plurality of dictionaries complicates the configuration and increases the cost.

An object of the present invention is to provide a character recognition device that can recognize handwritten characters containing a mixture of numbers, kana, alphabetic characters, etc., without error using a number of dictionaries.

In order to achieve the above object, the character recognition device of the present invention includes a dictionary that stores character bit patterns from which features of handwritten characters are extracted, and inputs video data of the read handwritten characters into a feature extraction and determination calculation section to extract the features. is extracted and compared with the dictionary, and a character type priority is given to the character obtained as a result of the determination operation, and a plurality of answer candidates are outputted, while a reading target character table is created that stores the character type of input handwritten characters. A final answer is obtained by inputting the plurality of answer candidates into a matching circuit, comparing them with the reading target character table, and outputting answer candidates that match the character types stored in the reading target character table. It is characterized by:

The present invention will be described in detail below with reference to examples.

FIG. 2 is an explanatory diagram showing the configuration of an embodiment of the present invention.

In the same figure, for example, a form 1 is inserted into the reading mechanism 2 of the OCR device 10 shown in FIG.
The stored video data is input into the feature extraction and judgment calculation section 11 shown in FIG. 2, where the features are extracted in the same way as in FIG. The first to nth candidates are output as answer candidates.

That is, the dictionary 6 stores character bit patterns of characteristics such as numbers, kana, alphabets, special symbols, etc., and compares them with characteristics of video data of input characters.

In this case, a priority is given to each feature item, and a plurality of answers 1 to An are output.

For example, if we were to give priority to numbers, kana, and alphabetic characters based on the feature content among the similar characters that were read, we would give first priority to the character closest to the feature of the number "7" mentioned above, followed by kana " The following priority is given to those showing a closer match to ``su'' or kana ``wa''.

Then, select multiple answer candidates with priority &1~A, n
A unique address is given to each of them and sent to the matching circuit 13.

On the other hand, a unique address is assigned to each of the characters to be read, such as numbers, kana, alphabetic characters, etc. According to these addresses, a character to be read table 12 is created, and the contents are sent to the matching circuit 1.
3 and is compared with the multiple answer candidates mentioned above.

That is, searching the reading target character table 12 corresponding to the unique address of the first priority character of the plurality of answers and detecting a match;
If they do not match, proceed to the second priority.

In the above example, if the characters to be read are the numbers "0-9" and the Japanese characters "a, i, tsu, nio", the number "7" is decisive, and there is no room for the Japanese character "su" to be output. .

On the other hand, if the only characters to be read are kana, the first priority digit "7" is discarded, and the second priority digit "su" and the third priority digit "wa" are replaced by "su". If the match with the dictionary is close, the second priority kana "su" will be output.

In the case of the number "8" and the alphabet rBJ mentioned above, it is decisive when either of the characters to be read is missing, but if both exist, the one with the closest match with the dictionary is selected as in the previous example. Either it will be given priority or both will be rejected as inconsistency.

As explained above, according to the present invention, a plurality of answer candidates are output by comparing the characteristic bit patterns of handwritten characters with the contents of a dictionary, while a reading target character table is created and compared with this to obtain a final result. The answer is:

As a result, there are fewer errors compared to the conventional method that judges the features of only handwritten character shapes and outputs only one answer, and the same effect as the above-mentioned plural dictionaries can be obtained with a single dictionary.

In this case, a reading target character table is required, but since this is a simple memory in which unique addresses are given to numbers, kana, alphabetic characters, etc., it does not pose much of a problem in terms of structure.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram of a conventional example, and FIG. 2 is an explanatory diagram showing the configuration of an embodiment of the present invention. In the figure, 1 is a form, 2 is a reading mechanism, 3 is a memory, and 6 is a dictionary (memory). , 11 is a feature extraction and determination calculation section, 12 is a reading target character table, and 13 is a collation circuit.

Claims (1)

[Claims] 1. A dictionary storing character bit patterns from which features of handwritten characters have been extracted is provided, and the video data of the read handwritten characters is input into a feature extraction and judgment calculation unit to extract the features and compared with the dictionary to obtain the result of judgment calculation. A character type priority order of the obtained characters is given to output a plurality of answer candidates, while a reading target character table is created that stores the character types of handwritten characters to be input, and the plurality of answer candidates are sent to a matching circuit. A character recognition device characterized in that a final answer is obtained by comparing the characters with the character table to be read and outputting answer candidates that match the character types stored in the character table to be read.