JPH03225489A - Contact character segmenting system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character reading device, and particularly to a method for cutting out characters one by one from a character string pattern.

[Conventional technology]

An example of a character recognition device that recognizes printed or engraved characters, etc. is as shown in FIG. It is quantized to about 8 bits and converted to binarization circuit 3.
The data is binarized and stored in the frame memory 4. A character cutting circuit 5 cuts out characters one by one from the pattern to be recognized stored in the frame memory 4. For example, a dictionary pattern previously stored in a dictionary pattern memory 6 and a recognition circuit 7
It compares and matches the patterns and outputs the most similar pattern as the recognition result.

In such a recognition device, if the quality of the characters is poor, the characters are likely to touch each other during the binarization process. Therefore, in order to separate the touching characters,
As shown in FIG. 7, a character cutting circuit 5 obtains a projection pattern (b) in a direction perpendicular to the rows of a character string (a), and cuts the projected pattern at a position where there are fewer projection patterns to perform character recognition. However, this method has a problem in that erroneous clipping occurs frequently and the recognition rate is extremely reduced.

For such simple processing, Japanese Patent Application Laid-Open No. 61-72373
No. 63-216188 and JP-A-63-216188 have been improved to prevent erroneous cutting, but since they both use a projected pattern, there are cases where the cutting position is incorrect for the contact pattern as shown in Figure 8. Ta.

In other words, since the projection pattern counts the number of pixels in the vertical direction, in the case of FIG. 8(a), the number of pixels in the vertical direction is the same for all portions A, so it is difficult to determine where in the portion A should be cut out. It is difficult to determine, and as shown in Figure 8 (b) -'
Because it was difficult to distinguish between the projection pattern of (hyphen) (section A) and the projection pattern of the upper part of '7' (section B), the cutting position could be incorrect. Furthermore, no consideration was given to how to deal with changes in character size or character spacing (character spacing is set at equal intervals).

[Problem to be solved by the invention]

Since the above-mentioned prior art determines the contact position using a projection pattern that focuses only on the number of pixels in the vertical direction, there is a problem in that it is not possible to distinguish -' (hyphen).

Further, there is a problem in that erroneous cutting occurs because no consideration is given to the case where the size of characters changes each time an image is captured.

SUMMARY OF THE INVENTION An object of the present invention is to provide a touching character extraction method that can correctly read character strings including touching characters.

[Means to solve the problem]

In order to perform character extraction taking into consideration (hyphen), the present invention uses the center X coordinate yc of the character string instead of the projection pattern.
(where the horizontal direction is the X coordinate and the vertical direction is the y coordinate) y. It is characterized by using a weight distribution such that pixels that are further apart have larger values. In addition, the character size (
It is characterized by determining the width) and cutting out based on this.

[Effect]

Focusing on the fact that the position of the hyphen (hyphen) is close to the character's center X coordinate '/c, and calculating the weight distribution around 'jc as described above, the pattern of pixel data close to 'jc will become smaller. , '7' has a large value because it is farther away than yc, and patterns that cannot be separated by mere projection distribution are not cut incorrectly.

In addition, for character blocks that cannot be separated even using the above weight distribution, find the standard character width and determine how many times this width the character block corresponds to.For example, if the width is three characters, then , by finding the position of three equal parts and cutting at the position where the weight distribution is the smallest in the vicinity of that position.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 2 shows the structure of the frame memory 4. As shown in FIG. If the horizontal direction is the X coordinate and the vertical direction is the y coordinate, 1 normal X is 256,
Dot pattern g (x t y
) is stored as image data.

Since the value of one dot, that is, one pixel is stored as a binary image, it takes the value of '0' or '1'.

The projection pattern mentioned in the prior art is the vertical accumulation of the number of pixels of a character, and as shown in Figure 3, the data of a certain pixel is g (x, y), and when it is a pixel of a character, g (
xty)=1 Otherwise, if g(x+y)=O, f(x)=Σg(x, j) is obtained for all X coordinates. Figure 3 (a
) is the character pattern (b) indicating its projection pattern.

The projected pattern f(x) of pattern (a) in which there is a hyphen ' (hyphen) between characters '8' and '7' in FIG. 4 is shown in (b). If a character is cut out at a portion where the value is larger than a certain threshold thl, part of '7' will be missing. Therefore, instead of this projection pattern, a weight distribution M (x ) around the center yc of the height of the character string shown in FIG. 3(a) is determined. This weight distribution M (x) is determined by the following equation.

M (x) = Σ(ly ycl ・g(x, y)
)...When the pixel of the character is '1', M(x)=Σ
(Iy-ycl ・g(x, y))...When the pixel of the character is '0', y is the X coordinate of each pixel with the upper left corner of the frame memory 4 as the origin . Also, the center height of the character string yc
For example, a projection pattern is obtained in the horizontal direction, and the midpoint of a region having a frequency greater than a threshold value is determined.

This M(x) is a value that increases as the distance from the center height yc of the character increases, so as shown in Fig. 3(c),
The difference between the - (hyphen) and the horizontal line above '7' is clearly visible. Therefore, character 1 ('8') and character 2 ('7') can be successfully extracted by cutting out characters using the threshold value thz.

By the way, when the characters in the character string in Figure 5(a) are separated using the above method, the parts that touch near the center height yc of the character string can be divided using the threshold value th. But character 1 ('5' and '2') character 5 ('5'
and '7') cannot be separated. Therefore, the overall processing is performed as shown in FIG.

First, the above-mentioned weight distribution M (x) is obtained by the 70-round weight distribution extraction circuit 8, and for this distribution, the cutting circuit 9 calculates the coordinates (up coordinates) of the portion larger than the threshold th, and Find the coordinates (dw coordinates) of the part. Further, the width of the character block cut out by the standard character width extraction circuit 10 is determined from the dw coordinate and the up coordinate, and the average character width is determined using only the character blocks satisfying a certain predetermined width. In the case of FIG. 5, characters 2, 3.4 are character blocks corresponding to this, and the average value of these character widths is determined. If there are few character blocks that satisfy the specified value (
For example, if the length is 2 blocks or less), the accuracy of the standard character width becomes poor, so it is estimated from the average height y4 of the entire character string. Letting the standard character width be X'A, it can be found as x-s=dXym (d is set in advance).

Then, a character block (character 1° character 5 in FIG. 5) which cannot be divided by the threshold value th with respect to the weight distribution M(x) is cut using the character division circuit 11. Specifically, how many characters of the standard character width corresponds to the width x4 of the character block is calculated using the standard character width x, -s as β=XdXX-
For example, if there are two characters, find the center X coordinate of
With respect to this
Divide at the X coordinate with (x). If it is three characters, you can separate the characters well by dividing it into three equal parts and repeating the same process.

With this method, even if the character spacing changes slightly, ±γ
(Even if the characters are printed at regular intervals, the character spacing will vary due to quantization errors, thickening or thinning of the characters, etc.)

By the way, the weight distribution M (x) is the center height of the character 'j
Since the calculation is performed around c, the cutting position occurs at the center of the alphabetic characters l Hl and 'N'.

If it is a number, all of them from 0 to 9 have a line on the top or bottom side, so they can be processed using the above-mentioned M (x), but alphabetic characters have the disadvantage of causing problems, but numbers can be reliably separated. These problems with alphabetic characters also occur in prior art methods using projection patterns.

〔Effect of the invention〕

As is clear from the above explanation, the cutting position of touching characters is determined using the weight distribution around the center height of the character string, so even if −′ (hyphen) touches a character, it can be divided well. Is possible.

Furthermore, even if the size of the characters changes, the dividing process is changed by determining the character width, so this can be handled well. Furthermore, even if the character spacing varies somewhat, it can be handled.

[Brief explanation of drawings]

FIG. 1 is a block diagram of the contact character extraction circuit of the present invention, and FIG.
The figure is a diagram explaining the frame memory, Figure 3 is a conceptual diagram of the projection pattern, Figure 4 is a diagram explaining the difference between the projection pattern and the weight distribution of the present invention, and Figure 5 is the weight distribution for character strings. Figure 6 is a block diagram of a character recognition device, Figure 7 is an explanatory diagram of a conventional character separation method using projection patterns, and Figure 8 is a diagram explaining problems with the conventional method. It is. 4... Frame memory, 8... Weight distribution extraction circuit,
9... Cutting circuit, 10... Standard character width extraction circuit, 1
1...Figure 2281m1j j; Figure 4 Figure 6

Claims (1)

[Scope of Claims] 1. In a recognition device that recognizes characters from a character string pattern, a weight distribution extraction circuit that determines the center height y_c of a character string and calculates a weight distribution in the vertical direction with this y_c as an axis;
An extraction circuit calculates the position where the distribution value is larger than a predetermined threshold value (up coordinate) and the position where it is smaller than a predetermined threshold value (dw coordinate), and cuts out the width of the dw coordinate and up coordinate as a character block, and from the character string. A standard character width extraction circuit extracts the standard character width, and the standard character width is used to determine how many characters the width of the character block touches, and when the width of the character block is two or more characters, the width of the character block is determined by the number of characters. A contact character cutting circuit comprising: a character dividing circuit which calculates coordinates equally divided by the number of characters, and calculates a position where the value of the weight distribution is the minimum value within a minute range of left and right with respect to the coordinates. method. 2. In claim 1, the center y-coordinate of the character height y_
The weight distribution M(x) around c is the pixel g(x, y) of the character line.
A contact character extraction circuit characterized by having a weight distribution extraction circuit that calculates the sum of the absolute values of the differences between the y-coordinate of Method. 3. In claim 1, the standard character width is defined as the average width of character blocks whose width satisfies a predetermined value determined by a cutting circuit, or the width of characters in a character string. A contact character extraction method characterized by having a standard character width extraction circuit that estimates from the height.