JPS5952473B2 - Character recognition method suitable for reading variable character sets - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention is an optical character recognition device (hereinafter referred to as OCR) in which characters to be read are numbers 0, 1, etc.
・・・9, Alphabet AB・・・Z, Kana Eye...
To provide a character recognition method that can effectively deal with cases that span a large number of character types (hereinafter referred to as categories) such as . It is.

As the uses of OCR become more diverse, it is no longer sufficient to target only numbers as in the past, and it is now required to be able to process alphabetical characters, kana, symbols, etc.

Naturally, an increase in the number of character sets means an increase in the types of characters that can be read. For example, if there are only numbers, there are 10 types, but if there are numbers and letters, there are 3 types.
There are 6 types of characters, numbers, alphabets, and kana sets, about 84 types, and if you add symbols, there are more than 100 types, which is 3 to 10 times more types than numbers only. Increasing the number of character types by 10 times not only means that the scale of the device (for example, storage capacity) and processing amount will be increased by 10 times, but also the identification itself becomes extremely difficult. This is because as the number of character types increases, the number of pairs of similar characters increases accordingly. For example, if it is just numbers, you can use pairs of similar characters such as "1" and "゛7" or "4" and "゛9", but if you have no English ability, you can use "0".
'Do゛D-゛゛1'' and 'I'', '゛゛2''Do'Z
Similar character pairs such as '', ゛゛8'' and ``B'' occur, and kana is added to ``1'' and ``do゛no''.
'゛2゛', ``Yu'' or ゛゛su'', ゛゛3'
' and 'yo', '4' and 'chi', the number '7'
The number of pairs of similar characters increases rapidly, such as ``'' and the kana ``ku''. Therefore, even when recognizing numbers that have the same shape, the level of difficulty will vary depending on whether you are recognizing only numbers, numbers and alphabets, or even numbers, alphabets, and kana. becomes very different. on the other hand,
OCR that can read numbers, alphabets, kana, symbols, etc.
When considering the usage pattern of ``<,'' it is rare that all categories are used together in the same document. Some documents contain only numbers, some documents contain only kana, and some documents contain only numbers and letters. It is often used as such.

Also, even within the same document, this field (character areas located in consecutive positions within a document is called a field) can contain only numbers,
In most cases, this field can be divided into fields such as alphabetic characters only, and kana characters only. Therefore, if there is a way to know what kind of character set and category the document or field currently being processed contains, it is possible to skip some processing by using that information. can. This is because, for example, the alphabetic character ゜“I” and the kana digit are not supposed to appear in the number field.
This is because there is no need to distinguish between 1r and ゛“pi or ``1゛ and ゜ノ゛.Also, in terms of readability, as explained earlier, the number of similar character pairs is reduced, so the character set is This can be expected to improve compared to unknown cases.For example, in the case of a document in which only numbers are used in the majority of fields, and numbers, alphabets, and kana are mixed in a small number of fields, the entire document is uniformly filled with numbers. , alphabetical characters, and kana, the overall reading rate is better if numeric fields are read using a method that only targets numbers, and mixed fields are read using a method that targets only mixed fields. It is obvious that this will happen.If we express it in concrete numbers, let's say that there are 200 characters (numbers) in the numeric field and 20 characters in the mixed field in one slip. If the reading rate per 0CR(7)l character to be processed is only numbers,
99%, and if numbers, letters, and kana are mixed, it is 95%. At this time, if the entire slip is read as a mixed field of numbers, letters, and kana, the reading rate is approximately 95%. However, if we use the field information, (200×99+2
A reading rate of 0×95)/220=98.6% can be expected. By using field information in this way, it is possible to increase the reading rate, but when using katakana in a certain field, for example, wo is used, and when using alphabetic characters mixed with numbers, it is difficult to read wo. D. I. S. If Z, etc. are not required, the reading rate can be further increased.

In these cases, because yo resembles wo, it will no longer be unreadable or misread as wo,
Similarly, since 8 is similar to B, 0 is D, 1 is similar to 41, 5 is similar to S, and 2 is similar to Z, it will not become unreadable or be misread.

Of course, in katakana, only the seven characters ゛, ku, ha, ni, ho, heto are used, or the English letters A. E. H.L. N. P. R
．． When using katakana or part of an alphabetic character set, such as when using only the 8 characters X, 46 katakana characters,
The reading rate is much better than when using 26 alphabetic characters, and even if it is mixed with numbers, there are fewer characters that resemble numbers, so the reading rate is lower than when using katakana and numbers, or alphabets and numbers. It gets expensive.

In this way, depending on how to register the character types in the character set,
A high reading rate can be obtained. The object of the present invention is to make the character set used for each field variable, and for recognition,
The purpose of the present invention is to provide a recognition method that eliminates unnecessary processing and increases the recognition rate. According to the present invention, there is a pattern memory that stores an input pattern for one character, a standard pattern memory that stores a standard pattern with a category name, and a character that stores in advance the character set name to which each character position on a slip belongs. Set name memory, Character set name register that specifies the name of the character set to which the input pattern belongs, Character set definition memory that indicates the categories registered in multiple character sets, Standard patterns read from the standard pattern memory means for determining whether the category of the standard pattern is registered in the character set specified by the character set name register by searching a character set definition memory; The present invention provides a character recognition method using a determining means that compares the standard pattern with the input pattern only when the character is registered in the specified character set, and determines the category of the standard pattern with the best match as the category of the input pattern.

Hereinafter, the present invention will be explained using the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention.

Reference numeral 9 denotes a scanning quantization unit that performs processing such as photoelectric conversion, cutting out, and preprocessing of characters; 10 a pattern memory that stores a pattern for one character; and 11 determines the position (or field) of an input character. The means 19 is a memory which uses the output of the input character position determining means 11 as an address and allows the character set name corresponding to the position of the input character to be read into the character set name register 16.

Reference numeral 14 denotes a standard pattern memory, which is composed of a category name of a standard pattern and characteristic conditions necessary for being determined as this category.

When the input pattern is input to the pattern memory 0, the determining means 22 reads one word at a time from the standard pattern memory 14 into the read register 15, and if the content is one input category, sets it in the category register 17. do. Character set name register 16 and category register 17
The address of the character set definition memory 20 is constructed by concatenating the .

The contents of the character set definition memory 20 are 1, assuming that there are 128 usable categories and 16 required character sets.
It has a capacity of 6 x 128 bits, or 2048 bits.
If this 2048-bit memory is composed of 1 bit and 1 word, the contents can be stored in character set name register]6 as 4.
By setting the bit and category register 17 to 7 bits, the data can be read to the register 21. That is, 2048 words, which is the capacity of the character set memory 20, are sequentially stored one by one.
It is divided into 16 units with 28 words each, and 128 categories are assigned to the 128 words to form one character set. 128 words contain numbers 0, 1, 2,...9
, English letters AB...Z, Katakana Aiu...
・・Wawon, the symbol +−・・・・・・・￥ are made to correspond in order, and ゜゜1゛ is stored in the word corresponding to the registered category.

When “0” is read in the register 21, the determining means 22 skips to the next standard pattern and writes “0” in the register 21.
1'' is read out, the characteristic conditions read out one after another from the standard pattern memory 14 to the readout register 15 are
It is checked whether the input pattern inside the pattern memory 1 river 0 is satisfied or not, and when reading of all the standard pattern memories is completed, the category name of the standard pattern that best satisfies the characteristic conditions is determined.

Let's explain using an example.

Figure 2 is an example of an 0CR slip, with the 0th line, 2nd line,
Columns 7 to 9 of the 3rd and 4th rows consist of either a number or a negative sign, and the first row consists of the alphabetic characters A. B. C. D, and the four characters in the 4th row, 1st to 4th columns are the alphabetic characters X. Y.
It is a mixture of Z and numbers, and all others are blank.

Space character set name 0 alphabetic character A. B. C. Only the 4 categories of D are character set name 1 alphabetic character X. Y. 2 character set names consisting of the 3 categories Z, and the numbers 0, 1, ...9. 3 character set names consisting of the numbers 0, 1, .... 9 and the negative sign (-). The character set name memory corresponding to the slip in FIG. 2 is as shown in FIG. 3, and the character set definition memory is as shown in FIG. 4.

In FIG. 4, the lines 0, 1, . . . 15 correspond to character set names, and the examples 0, 1, . . . 127 correspond to categories. 0, 1, ...9 columns are numbers 0, 1
, ~9, and the 10th to 35th columns correspond to the alphabetic characters A. Corresponds to B, ~Z, columns 36 to 81 correspond to katakana, and 82, 8
Columns 3 to 127 correspond to the symbols +, 1, △, . . . ¥.

For example, when the character ゜゜9゛ in the O row and second column in Figure 2 is input into the pattern memory, the character set name "゜゜゛" in the contents of the 0th row and second column in the character set name memory in Figure 3 is inputted into the character set. When the category name is read out from the standard pattern memory 4 to the readout register 15, the determining means 22 transfers the category name to the power lever name register 17, and transfers the category name to the character set definition memory 20. In the three lines of the character set definition memory in Figure 4, the category names are numbers 0, 1, ~9, and the negative sign is ゜゜1゛, so the standard pattern and input pattern for numbers and negative signs are Also, when the character ゜゜Y゛ in the 4th row and 2nd column in Figure 2 is entered into the pattern memory]0, the character set in the 4th row and 2nd column of the character set name memory in Figure 3 is compared. Name゜“2
゛ is read into the character set name register 16, and the standard pattern of numbers 0, 1, ~9 and alphabetic characters X, Y, and Z is stored with ``゜1゛'' in the second line of the character set definition memory in Figure 4. Matched against input pattern.

As described above, according to the present invention, it is possible to match the input pattern only when the standard pattern category is registered in the character set definition memory. It is possible to obtain 0CR with high recognition ability.

[Brief explanation of the drawing]

Figure 1 is a block diagram showing an example of the present invention, Figure 2 is an example of an OCR slip, Figure 3 is an example of the contents of a character set name memory, and Figure 4 is a character set definition memory. An example of the content is shown below. 9 is a scanning quantization means, 10 is a pattern memory, 11 is an input character position determining means, 14 is a standard pattern memory, 1
5 is a read register, 16 is a character set name register,
17 is a category register, 19 is a character set name memory,
20 is a character set definition memory, 21 is a register, and 22 is a determining means.

Claims (1)

[Claims] 1 A pattern memory that stores input patterns, a standard pattern memory that stores standard patterns with category names, a character set name memory that stores character set names in advance for each character position on a slip, and A character set definition memory in which a category is registered in a character set, and a character set read from the standard pattern memory in the character set specified at the character position by the character set name memory when determining the input pattern of the pattern memory. and a determining means for comparing the standard pattern with the standard pattern only when the category of the standard pattern that has been matched is registered, and outputting the category name of the standard pattern that best matches among the standard patterns that have been matched. A character recognition method suitable for reading variable character sets.