JPS59220887A - Normalizing method of size of character pattern - Google Patents

Abstract

PURPOSE:To prevent loss of information for identification of a character; to reduce the erroneous read of a character by adopting the normalizing method of the size of a character pattern where a part except a change point is thinned in case of reduction and a pattern is inserted to a change point in case of enlargement. CONSTITUTION:Row-direction and column-direction block number change detecting circuits 13 and 14 are provided additionally to an original pattern storage memory 12 where quantized character patterns are stored. The pattern is thinned except change points of block numbers of the stored character pattern which are detected by circuits 13 and 14, thus reducing the character pattern. Consequently, the important information for character recognition is prevented from being lost. Patterns are inserted to change points of block numbers of the character pattern to enlarge the character pattern, and thus, the important information for the character recognition is emphasized to reduce erroneous read for the character, read.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to an optical character reading device, and more particularly to a method for normalizing the size of a character pattern to correct variations in the size of characters to be read.

[Technical background of the invention]

When an optical character reader (OCR) (cCB) recognizes handwritten characters, it normalizes the size of the character pattern in the pre-processing stage of recognition in order to absorb variations in character size depending on the person writing the characters. .

This normalization of the size of a character pattern means that the character pattern is set to a predetermined constant size, that is, characters smaller than the standard are enlarged, and characters larger than the standard are reduced.

[Problems with background technology]

By the way, when reducing a character pattern, depending on the operation, important information for character identification may be lost, which may cause misreading. This is because conventional techniques for enlarging, reducing, and manipulating character patterns reduce the character pattern by partially thinning it out or enlarge it by inserting it according to a certain algorithm. In a reduction operation that thins out information in a prescriptive manner, the thinned out part may contain important information for character identification. This will lead to
This is a drawback that may lead to misreading.

[Purpose of the invention]

In view of the above-mentioned drawbacks, the purpose of this invention is to provide a method for normalizing the size of character patterns that prevents the loss of information necessary for character identification.

[Summary of the invention]

The invention is to transfer the quantized character information to the original figure pattern storage memory that stores the quantized character pattern for one character, and to determine the change point in the number of blocks in the row direction and column direction of the stored character pattern. By adopting a normalization method for the size of the character pattern, which detects ", in the case of reduction, thins out the part that avoids this change point, and in case of enlargement, inserts a pattern at this change point, the above purpose can be achieved. The goal is to achieve the following.

[Example of departure]

Hereinafter, one implementation of the inventive character pattern size normalization method U will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of an optical character reading device to which the uninvented method for normalizing the size of the character BUTTER/ is applied, and particularly shows the character recognition process. The characters written on the form 1 are read by an optical scanning circuit, and the characters written on the form 1 are converted into electrical signals by the optical conversion circuit 2, and the characters on the form 1 are converted into electrical signals.
Furthermore, this character signal is quantized. One line of the quantized character pattern on the form 1 is stored in the line buffer memory 3.The character detection and extraction circuit 4 extracts the character pattern from one line of the character image stored in the line buffer memory 3. Detect and cut out the character pattern for one character to be recognized.The cut out character is input to the next character normalization circuit 5, where it is converted into the quantized character pattern for one character manually created. The size is normalized. The character pattern whose size has been normalized is sent to the recognition circuit ll86. Here, it is compared with the dictionary memory 7, and the result of this comparison is sent to the judgment circuit 8, where the character is recognized. Get results.

FIG. 2 is an embodiment showing the detailed structure of the character normalization circuit shown in FIG. 1. The quantized character butter/signal for one character is manually input from the input bus 9 to the original graphic pattern storage memory 12 in the specified direction of the column address register 10 and row address register 11. This original pattern putter/storage memory 12
are connected to a row direction block number change detection circuit 13 and a column direction block number change detection circuit 14, and the original figure patterns detected by these row direction and column direction block number change detection circuits 13 and 14 are stored in the storage memory 12. A change detection signal for the block 1 of the character pattern is sent to the output box 15.

The character pattern in the original figure pattern storage memory 12 is normalized by a butter/gross sensor (not shown) by the method described later, and this normalized character pattern is stored on the specified surface of the column address register 16 and the row address register 17. , manually stored in the digitized pattern storage memory 18. The normalized pattern storage memory 18 outputs the compared character patterns to the forgetfulness circuit 6 shown in FIG.

31A shows an example of a character pattern stored in the original figure pattern storage memory 12 shown in FIG. 2, in which the katakana "a" is shown as one character. The code 31 is a register that holds the change point of the number of blocks in the line direction of the character pattern, and "0" indicates that there is no change in the number of blocks from the previous line.
// indicates that there has been a change in the number of blocks.

Reference numeral 32 is a register that holds the change point of the number of blocks in the column direction, and its operation is similar to that of the register 31.

Next, the operation of this embodiment will be explained. The character recognition process shown in FIG. 1 is the same as that described in the explanation of Bay Q, so the explanation of the operation will be omitted here.

Therefore, the operation of the character normalization circuit shown in the second (Numa) will be explained.
The character pattern cut out for each character is transferred from the line backer memory via the human power bus 9 to the address register 10.
11, the original figure pattern is stored in the memory 124. Here, for convenience, it is assumed that the character pattern stored in the original graphic pattern storage memory 10 is the katakana "a" pattern shown in FIG. The row direction block number tabulation detection circuit 13 sequentially searches the rows of the original figure pattern storage memory 10, and in the row where the number of blocks has changed, the corresponding both of the register 31 that holds the row direction block number change point are searched. If there is no change (1), set "0" in the column direction (the same goes for roughness). In the subsequent columns, SS 1 // or "0" is set in the register 32 that holds the change point of the number of blocks in the column direction.

Here, we will discuss enlarging and reducing character patterns for normalization. For example, the expansion is shown in Figure 3.
The size and height of the hawk-sized [A] in the row direction (also column direction))
is 18 bits.If you do this with 20 bits, you can make a 2-match, 2-bit row pattern (a row pattern is a pattern of 1'11 of the 3N) as a matchmaker. In this case, ・1 is placed at 1/3 and 2/3 of the character size.
-1 pattern is generally inserted. (In the case of 8th grade, the same idea is used to mechanically move certain places in both directions.
1-Tiyo'). At this time, Lelie is in the 3rd V! When the 10th row pattern is thinned out in the "A" pattern of 4,
As shown in the register 31, this part is t'l'r when the number of blocks is necessarily represented, and when this part is thinned out,
It is difficult to distinguish between Aj and ``F''. In the conventional character pattern reduction method, when reducing in the height direction, it is necessary to reduce the character pattern by n bits by 14 or 1.
This is to thin out fte butter/ at the location of 24 and . . . rVn+1 under adverse conditions.

As mentioned above, when the 10th row pattern in the pattern for "A" shown in Figure 3 is thinned out, it becomes difficult to distinguish between "A" and "N", and it becomes difficult to distinguish between "A" and "F". The information required for distinguishing characters is lost, leading to misreading. Therefore, in this embodiment, in view of the above points, the data in the row direction, column direction,
In the pronk number change detection circuit 13 and 14, the register 31
．． 32, ゝゝ1〃 is set in both places where the number of flocks of the character pattern is tabulated. , Therefore, in addition to normalizing the size of the character pattern, the putter knob 1 setter (not shown) calculates the location of the change point of the number of blocks when reducing by referring to the answers in registers 31 and 32. +Ichiゝ"1" is set)
The row pattern (or column pattern) of is not thinned out, but another location is thinned out and reduced.

On the other hand, in the enlargement method, the pattern processor preferentially inserts row patterns (or column patterns) at the part where the number of blocks changes and enlarges the pattern. In this way, the character pattern normalized by the pattern processor is stored in the normalized pattern storage memory 18 from the human bus 9 to the column address register 16 and the column address register 17 specified by the column address register 17.
It is stored in 1.

According to the seventh embodiment, the block number change detection circuits 13 and 14 in the row and column directions are attached to the 1st graphic pattern memory 12 that stores quantized character patterns, and these detection circuits @13. By thinning out the pattern by going to the change point of the number of blocks of the stored character pattern detected in step 14, it is possible to prevent important information from being lost in single character recognition because the character pattern is reduced to f17. Therefore, it is possible to reduce misreading by the character C-taking device. Also, to enlarge the character pattern by matching the pattern to a change point in the number of blocks of the character pattern.

Important information for character recognition is emphasized, and misreading by the character reading device can be reduced here as well.

〔Effect of the invention〕

As described above, according to the uninvented character pattern size normalization method, the quantized character butter/
In order to normalize the change points of blocks 2 and above of block 2 of the character pattern in the original figure pattern storage memory that stores them, in the case of reduction, the change points are avoided and thinned out, and in the case of enlargement, the patterns are inserted at the change points. This has the effect of preventing the 1n information for character identification from being lost. .

[Brief explanation of drawings]

FIG. 1 shows a configuration I of an optical character reading device to which the character pattern size normalization method of the present invention is applied.
A. FIG. 2 is a detailed configuration diagram showing one embodiment of the character normalization circuit shown in FIG. 1, and FIG. 3 is the original figure pattern storage memo 9 in FIG. FIG. DESCRIPTION OF SYMBOLS 1... Ledger, 2... Optical turtle exchange circuit, 4... Character exposure cutting circuit, 5 - Character normalization circuit, 12... Original figure pattern storage memory, 13... Change in the number of blocks between rows. Detection circuit, 14... Column block number change detection circuit, 1
8 ・Normalization pattern storage memory agent patent attorney rules
Kensuke Chika (and 1 other person) Figure 1 Figure 2 Figure 3 '1 0 1

Claims (1)

[Scope of Claims] 11J This is carried out to correct variations in the size of characters to be read when recognizing characters with an optical character reading device, etc.1. In the method for normalizing the size of a character pattern, the change points in the number of blocks in the row and column directions of the quantized character pattern for one character, which is the normalization stored in memory, are calculated, and this character pattern is When reducing / by mtj, reduce by thinning out the part that avoids the change point,
A method for normalizing the size of a character pattern, characterized in that when enlarging the character pattern, the character pattern is expanded by inserting a pattern at the change point. (2) As a method for detecting a change point in the number of blocks in the row direction and column direction of the character pattern stored in the memo 1 box, the memory is provided with a register that holds the point of converting the row direction block Mf and a column direction block Mf conversion point. A register is provided to hold the point, and 1" is set in the area of each register corresponding to the position where the number of blocks in the row and column directions of the character pattern changes. If there is no change, "0" is set. A method for normalizing the size of a character pattern according to claim 1.