JPS598864B2 - Pattern recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a pattern recognition device, particularly to a recognition method for an optical character reading device.

In optical character reading devices, input patterns are identified by matching the input pattern with a standard pattern, or by converting the input pattern into a feature code string and passing it through a sequential logic circuit. In some cases, an input pattern is accepted by two or more standard patterns and it is difficult to determine which category it should belong to.

Usually, such a pattern is rejected as a confusing pattern. That is, the category of the standard pattern closest to the input pattern is cm) the degree of similarity is smx)
The category of the next closest standard pattern is cm) the degree of similarity is Sm
If x, 5mx≧ε1, snx≧ε1 ゛゜゜゜゜゜゜゜゜0(1) 5mx-5nx≧ε2 ・・・・・・・・・
(2) However, the input pattern is determined to belong to the category Cm only after the threshold values of ε10 and ε2 are reached.

If 5mx-5nx<ε2 ・・・・・・・・・(3
), it will be rejected.

However, even if the pattern holds true for equation (3), it is often possible for humans to clearly distinguish the patterns. For example, in Figure 1, bw" and "ni" are difficult to distinguish for a character reader, but humans can recognize that the two left branches of "X" form an acute angle, and that of "ni" is approximately 1800. The two are distinguished by the way their branches intersect. In order to distinguish between such patterns, the present invention
Use the coordinates of both ends or one end of the branch.

When comparing the coordinates of both ends of a branch, which branch should be compared with which branch? Should the start or end of the branches be compared? Should the X coordinate be used or should the Y coordinate be used? varies depending on the patterns to be compared. Therefore, it is easier to compare if you know in advance what pattern the input pattern is similar to. For this purpose, the determination unit outputs the closest category Cm and the next closest category Cn, and compares the coordinates of the branches based on this information. Hereinafter, a specific embodiment of the present invention will be described with reference to the schematic configuration diagram of FIG. 2.

Characters written on paper are converted into electrical signals by a photoelectric conversion device and sent to a preprocessing device.

The preprocessing device removes noise, normalizes the size, normalizes the position, and cuts out each character. Each character pattern is reduced to a line width of 1 bit by a reduction circuit and converted into a line figure. Feature points such as end points and branch points are extracted from this line figure. This table will hereinafter be referred to as a feature point table. On the other hand, a feature code string having each feature point as a starting point and an ending point is created from the line figure by a feature encoding circuit.
Each branch consisting of this feature code string is drawn in the form of a table containing the coordinates of the start and end of the branch, the length of the branch, and the like. This table is hereinafter referred to as the branch table. In this embodiment, an example is given in which an input pattern is narrowed into a line figure and feature points and a feature code string are obtained from the line figure, but the feature extraction circuit can be designed in various ways to meet the above requirements.

Other feature circuits may be used as long as the feature points at both ends of the branch can be found by some method. If the input pattern is expressed as a feature code string for each branch, matching is performed by making the standard pattern and input pattern correspond to each other for each branch.

Cm the standard pattern category closest to the input pattern
When equations (1) and (2) are satisfied, the input pattern is determined to belong to the category Cm, with the standard pattern category closest to Smxl having the l similarity and Smx having the Cnl similarity. If equation (3) is satisfied, the category pair Cm, Cn is output and sent to the coordinate comparison circuit. The coordinate comparison circuit finds a branch to be compared based on the category pair Cm and Cn, and determines which category the input pattern belongs to. The branches to be examined differ depending on the category pair Cm and Cn, so remember them in advance in a table.

Or CmCn without showing it all in hardware.
It is also possible to input the signals to the machine circuit by a combination of the following.

For the sake of simplicity, an example will be described in which pairs of branches to be examined based on category pairs are stored in the form of a table.

FIG. 3 shows a parameter table containing category pairs, branch numbers, threshold values, etc. In this figure, ClC,
Columns indicate category pairs. In this figure, branch numbers are attached to patterns indicating categories. The I,,I, column stores the number of the branch to be compared. The TIT column stores threshold values. The input pattern from the matching circuit is ▲
~) 6K", the term "XK" is found from the parameter table and the second and third branches are known to be compared.

This branch will be compared using the coordinates shown in FIG. The start and end of the second branch are (XrYT), (X?, Y5),
The third start and end are (Xt, Yr), (X stroke Meng)
Then, it is determined that (X: -

The above operation will be explained in more detail using the coordinate comparison circuit shown in FIG.

The determination circuit sets the category pair CmCn in the register 1r.

The address register 10 is set at the beginning of the parameter table, read out in order, and set in the memory register 13. C
A comparator 18 compares lC, and CmCn. If they are not equal, adder 11 is operated with signal 100, the contents of address register 10 are incremented by 1, and the next information is read from parameter table 12. If ClC, and CmCn are equal, the contents of 11 are set in the address register 14. The contents of the branch table are read into the register 16 using the register 14 as an address. In this figure, for simplicity, only the coordinate information is read out. register 1
6, the starting end (X, Yl) and ending end (X, Y,) of the branch are read out. It is determined whether the X coordinate or the Y coordinate is to be used using signals 103, 104, 105, and 106, which are input to a subtracter 19. First subtractor (X1-X
, ) are set in register 20.

Whether the data should be entered into register 20 or register 21 is controlled by signals 101 and 108. Next signal 1
At 02, the contents of 1, are transferred to the address register 14, and 1,
The branch table of the th branch is read into the register 16. This time it is stored in the register 21 in the same way as before. Next register 2
0 and the contents of register 21 are added and stored in register 23. The register 23 has (heart-XJ)++(X1-
XI) will be included. This content is compared with the content of T2 using a comparator 24. If (heart-X1!) + (
If Xi−X2j)≧T, the contents of Cm are transferred to the register 26 by a signal 109.

If (Min - Move to.

If (X, l single mind) + (X{-XJ)≧T, If so, reject the input pattern.

Through the above processing, the result determined by the coordinate comparison circuit is obtained in the register 26.

In the above example, the difference in the X coordinates at both ends of the two branches was compared, but it is not necessarily necessary to use two branches, nor is it necessary to limit it to the X coordinate.

It is also possible to use both the X and Y coordinates, the coordinates of one end of a single branch, or the coordinates of three or more branches. In any case, by using the present invention, even a pattern that has conventionally been rejected as a misleading pattern can be correctly recognized by comparing the coordinates of the branches.

Note that since the branches to be compared are limited using category pairs, there is no need for the redundant process of comparing the coordinates of all branches, making it possible to simplify the hardware and improve recognition accuracy.

[Brief explanation of the drawing]

Figure 1 shows an example of a confusing pattern pair, Figure 2 shows an example of a confusing pattern pair;
3 is a diagram showing the contents of a parameter table, FIG. 4 is a diagram showing how to take the coordinates of both ends of a branch, and FIG. 5 is a diagram showing a coordinate comparison circuit. FIG. 12 is a storage device for storing a parameter table, and 10 and 13 are its address register and memory register, respectively. 15 is a storage device in which the branch table is stored, and 14 and 16 are its address register and memory register, respectively. 1T, 20, 21, 23, 26 are registers, 18, 24
, 25 is a comparator, 19 is a subtracter, and 22 is an adder.

Claims (1)

[Claims] 1 Calculate the similarity between the input pattern and the standard pattern,
A first determination step of determining which standard pattern the input pattern closely matches, and when a plurality of the standard patterns that closely match are obtained through the first determination step, the input pattern matches the standard pattern. In a pattern recognition method that includes a second determination step of re-determining which of the patterns it belongs to, the second determination step creates a branch table that defines the branches of the input pattern, and then determines which of the standard patterns it belongs to. By quoting the coordinates of both ends or one end of a branch in a predetermined part from the above branch table,
A pattern recognition method comprising the step of comparing the input pattern with each of the standard patterns and performing the re-judgment.