JPS598863B2 - Pattern recognition method - Google Patents

Description

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

In optical character reading devices, the input pattern is 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 either method, the input pattern or two or more standard patterns may be accepted, making it difficult to determine which category the pattern should belong to.

Usually, such a pattern is rejected as a confusing pattern. In other words, 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 is smx≧ε1, snx≧ε, ...
・・・・・・・・・・・・・・・(1) 5mx-5nx
≧ε2 ・・・・・・ ・・・・・・・・・・・・・・・
(2) Here, the input pattern is determined to belong to the category Cm only after ε4 and ε2 reach the threshold values.

5mx-5nx<ε2 ・・・・・・・・・ ・・・・
...... In the case of (3), this pattern is rejected as a misleading pattern.

However, (even if the pattern consists of three elements, it is often possible to clearly distinguish whether it is a person or not. For example, ゛J"bs'' shown in Figure 1 is a pattern that is difficult to distinguish based on similarity alone. However, while the legs of the person "bi" are bent,
It is known that the legs of "s" are straight and can distinguish between the two.The present invention utilizes the directional code at both ends or one end of the branch to distinguish between these patterns.

In other words, the branches that make up the legs of "bi" start at its starting point or at the bottom right, while the branches that make up the legs of "T" start at its starting point or at the top. When comparing the direction codes of both ends or one end of a branch, which branch should be compared differs depending on the input pattern. For example, when distinguishing between "J" and "T1", it is sufficient to compare the branches of the leg, but "B" and "8" shown in Fig.
Now we have to compare the direction of the end of the first branch and the direction of the start of the second branch. (However, the order of the branches is assumed to be in the numerical order shown in the figure.) However, it is easier to compare if you know in advance which pattern pair the input pattern is close to. Therefore, the discriminator outputs the closest category Cms and the next closest category Cn, and based on this information, the branch to be compared is determined. Hereinafter, specific embodiments of the present invention will be described with reference to the figures. I will explain.

FIG. 3 is a schematic configuration diagram showing a pattern recognition device of the present invention. 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 extracts each character. Each character pattern is converted into a line figure with a line width of 1 bit by a thinning circuit. From this line figure, feature points such as end points and branch points are extracted by a feature point extraction circuit. On the other hand, a direction coding circuit generates a direction code string for each branch from the line figure, with each feature point as the starting point and ending point. The direction encoding circuit is a circuit that sequentially represents the direction of a minute line element by an eight-directional code shown in FIG. 4, starting from the starting end, and converts it into a code string along the line to the terminal end. This circuit also outputs information such as the coordinates of both ends of the branch and the length of the branch in the form of a table. This table is hereinafter referred to as the branch table. If the input pattern is expressed as a direction code string for each branch, matching is performed by making the standard pattern and the input pattern correspond to each other for each branch.

Cm the standard pattern category closest to the input pattern
l Similarity is Smxl The category of the standard pattern closest to Cn similarity is Snx (1), (If the two conditions are satisfied, the input pattern is determined to belong to the category Cm〇 If the equation (3) When satisfied, the category pair Cm, Cn is rejected and sent to the average direction code comparison circuit as an output for further detailed investigation. The direction code comparison circuit calculates the branch to be compared using Cm, Cn, and calculates the branch to be compared based on the input pattern. Determine which category belongs to 〇 The branches to be examined differ depending on the category pair Cm, Cn, so make a table of them in advance and memorize them. Or, to make all of them into hardware, Cm, The signals may be input to different circuits depending on the combination of Cn.

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. 5 is an example of such a table. Hereinafter, this table will be referred to as a parameter table.Category pairs are stored in the Cl and C2 columns of this table. In FIG. 5, the numbers 1, 2, etc. in the figures in this column indicate the order of the branches. The 1112 column stores branch numbers.

The NlN2 column stores the length of the code to be averaged. The FlF2 column stores information that distinguishes between the starting end and the ending end of a branch. If Fi = 1, it indicates the starting point, and if Fi = 0, it indicates the ending point. 〇 Use this table to indicate the number of the branch to be compared,
Find threshold values, etc. For example, if the determination circuit outputs category pair 7JT7, the starting end direction of the first branch is C.
Similarly to s, find the terminal direction CE of the first branch and calculate CS-
If CE≧T2, then 7J'' and C8-CE<T1, then 4T'' is determined.Otherwise, it is rejected.Let us explain in more detail the processing procedure above.

FIG. 6 is a schematic diagram of the average direction code comparison circuit. This circuit consists of a parameter table reading circuit, a branch table reading circuit, an average direction calculation circuit, and a comparison circuit. FIG. 1 shows a parameter table reading circuit and a category pair comparison circuit.

When the category pair CmCnP is output from the judgment circuit,
Set in register 14. The contents of the address register 10 are set to the first address of the parameter table, and the information is read out in order from the parameter table 12 to the memory register 13.
Store in. A comparator 15 compares ClC2 and CmCn. If not equal, adder 11 and register 10
Add 1 to the contents and read the next information. If they are equal, the contents of the branch table are read out using signal 101. FIG. 8 shows a branch table reading circuit. 0 First, the contents of I, are set to the address register 20 using the signal 101. 0 The contents of the branch table are read at this address and set to the register 22. 0 For simplicity, the contents of the branch table are set to the register 22. Assume that only the branch length L and the start address ADR of the direction code string are read out.〇A
The contents of DR are put into the address register 23 and the direction code string is read out into the register 25.

FIG. 9 shows an average direction calculation circuit.

To find the average direction code, calculate .

First, the contents of register 25 are put into shift register 31.

When F1 is O, the code is shifted to the left by L-N1. Assuming that one code consists of 4 bits, the arithmetic unit 33 multiplies it by 4 and sets the branch to be shifted in the counter 34. Each time a clock is added, the contents of the counter are decremented by 1, and the game stops when the number is one. When F1 is 1,
No shifts will be made. This means that the first code CO of the codes to be averaged has arrived at the beginning of the shift register 31. Note that the contents of N1 are set in the counter 30. First, CO is set in register 36.

Next, the counter 30 is decremented by 1, the counter 35 is set to 4, and the 4-bit shift register is shifted with a 0 tarlock.
C1 is also brought to the beginning of the shift register 31. Set C1 in register 31. CO-C1 is set in register 39 using 38 subtracters. Next, register 39
The contents of and the contents of register 41 are added and stored in register 41. However, it is assumed that the register 41 is initially set to 0. Furthermore, set the counter 35 to 4,
Take the following code and perform the same calculation until the counter 30 is O.
Continue until. If it becomes O, the contents of the register 41 are divided by N, and the result is stored in the register 43. Furthermore, the contents of the register 36 are added thereto using the adder 44, and the result is stored in the register 45. This means that the average direction code for the 11th branch has been compiled. Then 1 at signal 102
2 is set in the register 20, and the average direction code of the 12th branch is obtained in the same manner and stored in the register 46 this time.

Next, the contents of the register 46 are subtracted from the contents of the register 45 and stored in the register 48. This contents and T2 are compared to the comparator 4.
If C8-CE≧T2, the contents of Cm are transferred to the register 51, and if not, then the comparator 50 compares the contents with T.

CS-CE<T1 If so, the contents of Cn are moved to register 51.

Otherwise, a reject signal 103 is issued. The register 51 stores the category name determined by the average direction code comparison circuit. In the above example, the difference between the direction codes of one or two branches is determined and compared with the threshold value, but there are cases where the difference does not necessarily need to be determined.

The direction of the starting end or end of one or several branches may be compared, or the relationship between the directions of three or more branches may be compared. In any case, by using the present invention,
Even patterns that were conventionally rejected as confusing patterns can be correctly recognized by comparing direction codes.

Furthermore, since the branches to be compared are limited using a parameter table, there is no need for the redundant process of comparing the direction codes of all branches, which simplifies the hardware and improves recognition accuracy. In the present invention, the direction code is used, or when a candidate category is selected,
A portion of the candidate categories with notable differences in features may be determined in advance using other parameters (other than the direction code), and this may be used as the criterion.

[Brief explanation of the drawing]

Figures 1 and 2 are diagrams showing confusing pattern pairs, Figure 3 is a schematic configuration diagram showing an embodiment of the present invention, Figure 4 is a diagram showing an 8-direction code, and Figure 5 is a parameter table. FIG. 6 is a schematic diagram of the average direction code comparison circuit, and FIG. R is a diagram of the parameter table reading circuit and category pair comparison circuit. 12 is a storage device storing a parameter table; 10 and 13 are respective address registers and memory registers. 14 is a register, and 15 is a comparator. FIG. 8 is a branch table reading circuit diagram. 21 is the branch table, 2
4 is a storage device that stores the direction code string. 20 and 23 are address registers, and 22 and 25 are memory registers.

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 judgment step of re-judging which of the patterns it belongs to, the second judgment step is a pattern recognition method that includes a second judgment step of re-determining which of the patterns the pattern belongs to. Compare the input pattern with each of the standard patterns by using the direction of the branch,
A pattern recognition method comprising the step of re-judging as described above.