JPH05290169A - Fuzzy pattern recognizing device - Google Patents

Abstract

PURPOSE:To provide the fuzzy pattern recognizing device which evaluates an important feature part of an input pattern and reflects it on the final degree of recognition confidence. CONSTITUTION:A 1st main processor 3 compares a read image which is converted by a signal conversion part 2 into binary data with a pattern (fuzzy templet) for partial evaluation to calculate the degree of confidence of recognition for the fuzzy templet. A 2nd main processor 4 compares the read image with a fuzzy templet for total evaluation to calculate the degree of confidence of recognition for the fuzzy templet. A 3rd main processor 5 weights the degrees of confidence calculated by the 1st and 2nd main processors 3 and 4 with coefficients (1-alpha) and a to calculate the final degree of confidence.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fuzzy pattern recognition apparatus having a function of accurately recognizing a pattern transformed from a basic pattern, such as a handwritten character for a typeface, by using a fuzzy theory.

[0002]

2. Description of the Related Art A conventional pattern recognition device generally recognizes a pattern whose shape is strictly defined.
A method using a neural network is known as a means for recognizing an ambiguous pattern that is not limited to a fixed shape such as handwritten characters. in this way,
A neural network is formed by learning the pattern to be recognized in advance, and the input pattern is recognized thereby.

However, the method using the neural network requires a work of learning, and requires re-learning every time the number of pattern candidates to be identified increases. In addition, if the neural network is a black box and changes in the characteristics of candidate patterns to be recognized due to aging of the input device that captures the pattern signal to be recognized, the changes are reflected in the operation of the device. However, there is a problem that erroneous recognition may occur.

On the other hand, in recent years, a method of recognizing an ambiguous pattern using a fuzzy theory has been proposed. This method is based on having a plurality of knowledge about the characteristics of the pattern and checking them one by one, such as "there is a vertical line near this portion". However, in the conventional pattern recognition method based on the fuzzy theory, it is necessary to acquire knowledge about pattern features (rule creation work) when developing a system including software, and pattern feature extraction is performed when actual pattern recognition is performed. Algorithm is needed. for that reason,
There is a problem that processing is complicated because many rules and membership functions of fuzzy sets are required.

[0005]

Therefore, in order to solve the problem of pattern recognition by the set of the conventional fuzzy theory, the inventor of the present application recognizes the combination of the membership functions of preset fuzzy sets. Prepare a basic pattern to be matched with the target and compare the pattern with the recognition target pattern to calculate the certainty factor, which is an index for evaluating the pattern. (Japanese Patent Application No. 2-27)
4355).

According to this method, a pattern to be collated with a recognition target is formed by combining membership functions indicating a fuzzy region having a boundary of an arbitrary shape (for example, an ellipse). It is possible to efficiently recognize patterns such as characters and figures without requiring troublesome work. The matching pattern formed by combining the membership functions by this method is called a "fuzzy template", and this "fuzzy template" is also used in this specification.

By the way, for example, the numbers "0", "2",
When recognizing the number "2" from a limited group of recognition target patterns such as "4", "6", and "8", as shown in FIG. 14 (a) empirically or experimentally. It is necessary and sufficient to identify the characteristic portions 21 and 22 that do not exist in other numbers. Similarly, when recognizing the number "4",
It is necessary and sufficient to identify the characteristic portions 23 and 24 as shown in FIG. In addition, the characteristic part 2 of the number "2"
1 is a diagonal line and another feature 22 is a lower horizontal line. The characteristic portions 23 and 24 of the numeral "4" are vertical lines and horizontal lines which are orthogonal to each other.

However, the above-mentioned characteristic portions 21 to 24
When a line segment other than is added to the fuzzy template as a recognition target, in a situation where the input pattern is easily disturbed by noise or the like, when the fuzzy template of the number “2” is collated with the input pattern of the number “2”, There is little difference between the certainty of recognition and the certainty of recognition when other fuzzy templates are matched. Therefore, for example, in a fuzzy decision support system that determines operations such as control based on the recognition result, there is a problem that the situation cannot be clearly determined. Further, depending on how the input pattern is disturbed, an erroneous recognition result may occur due to the evaluation result of a portion other than the necessary and sufficient characteristic portion.

In view of the above problems, it is an object of the present invention to provide a fuzzy pattern recognition apparatus capable of evaluating an important characteristic part of an input pattern and reflecting it on the final recognition certainty factor. ..

[0010]

According to the present invention, in a fuzzy pattern recognition device for performing pattern recognition based on a membership function representing a fuzzy set, a pattern input section for inputting a pattern to be recognized and the membership function are provided. By generating one or more and combining the membership functions, an overall evaluation pattern and a partial evaluation pattern for recognizing the entire and characteristic portions of the recognition target pattern input through the pattern input unit are formed. A matching pattern forming means and an arithmetic processing means for selectively performing an operation for calculating the certainty factor of recognition by comparing the entire and characteristic portions of the pattern to be recognized with the overall evaluation pattern and the partial evaluation pattern, respectively. It is characterized by

In the present invention, when the certainty of recognition is calculated by comparing the characteristic part of the pattern to be recognized with the partial evaluation pattern, the partial evaluation pattern is the region of the characteristic part of the pattern to be recognized. It is preferable that it can be set only to. Further, the arithmetic processing means compares the entire and characteristic portions of the pattern to be recognized with the overall evaluation pattern and the partial evaluation pattern, respectively, to calculate a certainty factor of recognition, and weights this certainty factor to make a final recognition. It is preferable to calculate the certainty factor.

[0012]

In the present invention, the arithmetic processing means selectively performs the arithmetic operation for calculating the certainty factor of the recognition by comparing the entire and characteristic portions of the pattern to be recognized with the overall evaluation pattern and the partial evaluation pattern, respectively. .. Therefore, when the input pattern is not disturbed by noise or the like, the entire recognition target pattern is compared with the overall evaluation pattern to calculate the recognition certainty, and when the input pattern is disturbed by noise or the like, Can compare the characteristic part of the pattern to be recognized with the partial evaluation pattern and calculate the certainty factor of the recognition to evaluate the important characteristic part of the input pattern.

Further, the entire and characteristic portions of the pattern to be recognized are compared with the overall evaluation pattern and the partial evaluation pattern, respectively, to calculate the recognition certainty factor, and the certainty factor is weighted to obtain the final result. It can be reflected in a certain recognition certainty.

[0014]

1 is a block diagram showing an embodiment of a fuzzy pattern recognition apparatus according to the present invention. The device is configured to recognize characters and consists of the following components.

The image reading apparatus 1 includes a light source for illuminating a recognition target in order to read an image of a typeface or handwritten characters,
It has a photoelectric converter which reads an image by reflected light of the image and converts it into an electric signal. Instead of this, a pressure-sensitive handwritten character input device may be used.

The signal converter 2 converts the image signal read by the image reading device 1 into a binary signal of, for example, "0" (= "white") or "1" (= "black"), and , Is configured to correct (scale) the size of the read image. In this correction, for example, when a character has a vertically squeezed shape, the shape is vertically stretched so as to match the aspect ratio of a collation pattern (fuzzy template) described later.

The image reading device 1 and the signal conversion section 2 constitute a pattern input section for inputting a recognition pattern.

Next, the first main processor 3 compares the read image binarized by the signal conversion unit 2 and the shape of which has been corrected with the partial evaluation fuzzy template to recognize the fuzzy template. Calculate confidence.

The second main processor 4 compares the read image with the fuzzy template for overall evaluation to calculate the certainty factor of recognition for the fuzzy template.

As will be described later, the third main processor 5 weights the two certainty factors calculated by the first and second main processors 3 and 4 with coefficients (1-α) and α, respectively, Calculate the final certainty factor. The main processor 5 recognizes the read image as a fuzzy template pattern having the maximum final certainty value, and outputs the shape number and the shape name.

The above-mentioned main processors 3 to 5 constitute an arithmetic processing means for selectively performing an arithmetic operation for calculating the final recognition certainty factor. Although these processors are shown as individual constituent elements in the figure, they are composed of one processor (CPU) as hardware.

Next, the storage device 11 is composed of the following storage units 6 to 10.

The partial evaluation algorithm storage unit 6 stores an algorithm for the first main processor 3 to evaluate only the characteristic part of the input pattern, and the partial evaluation fuzzy template storage unit 7 stores this part. Fuzzy templates used for dynamic evaluation are stored. In addition, in the overall evaluation algorithm storage unit 8, the second
An algorithm for the main processor 4 to evaluate the entire input pattern is stored, and the fuzzy template storage unit 9 for general evaluation stores a fuzzy template used for this general evaluation. Further, the final evaluation algorithm storage unit 10 stores an algorithm for the third main processor 5 to perform a final evaluation by weighting partial and overall evaluations.

Each fuzzy template stored in the partial evaluation fuzzy template storage unit 7 and the overall evaluation fuzzy template storage unit 9 is, for example, a CRT used as a terminal device of a computer and a man-machine interface 13 such as a keyboard. Can be adjusted and changed via. The weighting coefficient α stored in the final evaluation algorithm storage unit 10 can also be changed via the man-machine interface 13 in the same manner.

The three-dimensional membership function generator 12 is
The three-dimensional membership function necessary for setting the fuzzy template as described above is generated.

The storage device 11, the three-dimensional membership function generator 12 and the man-machine interface 13 are
It constitutes a matching pattern forming means.

The signal converting section 14 outputs the pattern as the final recognition result by the third main processor 5 to 2
The digitized signal is converted into an analog signal and sent to the output device 15 such as a CRT.

Next, the operation of the embodiment will be described with reference to FIG.

First, when an image signal is input from the image reading device 1 (step S1), the signal conversion unit 2 performs preprocessing. That is, the image signal is converted into a binary signal, and necessary correction (scaling) is performed (step S2).

In the first main processor 3, the read image binarized by the signal conversion unit 2 and the shape of which is corrected and the fuzzy template for partial evaluation are compared, and the certainty factor of recognition for the fuzzy template is calculated. Partial evaluation is performed (step S3).

Further, the second main processor 4 compares the read image with the fuzzy template for overall evaluation, and carries out overall evaluation for calculating the certainty factor of recognition for the fuzzy template (step S4).

The third main processor 5 calculates the final certainty factor by weighting the two certainty factors calculated by the main processors 3 and 4 with the coefficients (1-α) and α (step S5). Specifically, when α = 0, the partially evaluated certainty factor is calculated as the final certainty factor, and α =
In the case of 1, the overall reliability is calculated as the final reliability. The signal indicating the finally recognized result is post-processed by the signal conversion unit 14. That is, the binarized signal is converted into an analog signal and sent to the output device 15 such as a CRT (step S6).

Next, the principle and method of pattern recognition in the embodiment will be described.

First, one or more three-dimensional membership functions are combined to create a fuzzy template corresponding to the typeface of the basic pattern to be recognized. The three-dimensional membership function used for this purpose will be described with reference to FIGS. 3 and 4.

(A) Setting of Membership Function For example, consider a membership function having an elliptic equi-fitness curve as shown in FIG. 3 in a three-dimensional space consisting of xy Cartesian coordinates and goodness of fit. Here, the symbols in the figure are defined as follows. For simplicity, the origin is the reference point (the center point for describing the 3D membership function most easily).
And

F (x, y, R _x , R _y ) = 0: Function giving the elliptical shape of the lines of equal fitness t _x = f (x): Bell type membership function t _y = g (in the xt _x plane y): Bell-shaped membership function on the yt _y plane R _x : Radius in the x-axis direction of the elliptic iso-fitting line with goodness of fit “1” R _y : Elliptical equality with goodness of fit “1” Radius in the y-axis direction of the parallel line a _x : Parameter proportional to fuzzy entropy about x a _y : Parameter proportional to fuzzy entropy about y r _x : including arbitrary point (x, y), etc. Distance between the goodness of fit line and the goodness of fit line of "1" on the x-tx cross section r _y : The goodness of fit line including the arbitrary point (x, y) and the goodness of fit of "1" Distance t on the yt _y section from the degree line: Fitness of the point (x, y) given by the three-dimensional membership function For convenience of explanation, the fitness t is divided into t _x and t _y .
t, t _x and t _y are virtually the same coordinate axes.

At this time, the iso-fitting line including an arbitrary point (x, y) is as follows.

[0038]

[Equation 1]

The three-dimensional membership function for t _x and t _y is

[0040]

[Equation 2]

[0041]

[Equation 3]

Is represented by Here, − of decoding is x, y
Indicates a positive side portion, and + indicates x and y negative side portions.

On the xt _x cross section and the yt _y cross section, the distances between the equal conformance line of interest and the reference point are R _x + r _x and R _y + r _y , respectively.
Because,

[0044]

[Equation 4]

[0045]

[Equation 5]

Therefore, from these,

[0047]

[Equation 6]

At this time, unless the denominators of the first and second terms have the same shape in parentheses, they cannot be transformed into an explicit function of t. _{Therefore, min (R x / a x} , R y / a y) = s k seek ... (7) to select the value of the appropriate s from a range of 0 <s <s _k,

[0049]

[Equation 7]

_Then , d _x and d _y are obtained. That _{d x = R x -a x s} ···· (9) d y = R y -a y s seeking (10), deforms the function equal fit line as follows.

[0051]

[Equation 8]

Here, in the decoding, − is the part where x and y are on the positive side,
+ Represents the part where x and y are on the negative side.

[0053]

[Equation 9]

The geometrical meaning of this equation represents a shape approximated by adding a linear portion to the original elliptical shape as shown in FIG. This formula gives

[0055]

[Equation 10]

However, when | x | <d _x , x = d _x , and when | y | <d _y , y =
Calculate as d _y . In addition, as shown in FIG. 3, when the focused equi-fitness line is outside the ellipse of the fitness “1”, {}
The inside is a positive value.

In the same manner, the same formula can be obtained when the target equi-fitting line is inside the ellipse having the fitting degree "1". However, in this case, when any point (x, y) exists in the straight line part of the approximate equi-fitting line, | x | <d _x and | y | <d _y, and the inside of {} is positive. And the value inside {} is a negative value when it is in the elliptical part.

When the inside (or the outside) of the ellipse having the goodness of fit "1" is a region having the goodness of fit "1" uniformly, a value is given by judging according to the above condition.

As an example, "8" as shown in FIG.
The fuzzy template represented by the three-dimensional membership function for the character graphic of is as follows from Eq. (14).

When the range of the goodness of fit of the three-dimensional elliptic membership function having the goodness of fit of “1” is the circle having the center point of (6,14) and the radius of 3.5 (see FIG. 6). )

[0061]

[Equation 11]

When the range of the goodness of fit of the three-dimensional elliptic membership function having the goodness of fit “1” is the circle having the center point of (6,6) and the radius of 4.5 (see FIG. 7). )

[0063]

[Equation 12]

As shown in FIG. 8C, the following function obtained by the combination of the above-mentioned three-dimensional elliptic membership functions is taken as the three-dimensional membership function of the fuzzy template for the above-mentioned character graphic "8".

T = max (t ₁ , t ₂ ) ... (17) Here, the reason why the range of the goodness of fit is expanded to [-1,1] is that the input coordinate values are inappropriate. This is because an evaluation value of- (minus) is given, and for example, when a completely black image signal is input, the total sum of the degree of conformity described later becomes a small value.

(B) Conversion of image input The input image is read by the image reading apparatus 1, and the signal conversion unit 2 performs conversion into a binary signal including size correction.

To explain using FIG. 5A as an example, in this case, as shown in FIG. 5B, there are 12 regions including the character "8" in the x-axis direction and 19 regions in the y-axis direction. It is divided into individual areas, and each area is converted into a binary signal corresponding to the character shape.
Here, the blank area is converted into a binary signal “0”, and the black area is converted into a binary signal “1”. The size correction is
The second black areas from the upper, lower, left, and right ends are located in the second row from the upper, lower, left, and right ends.

At this time, as shown in FIG.
The binary signal "1" (black) is given to the area of n = 35 out of the area of 19 = 228.

(C) Calculation of fitness of input image With respect to the membership function t set in (A), “n” of the n black regions (binary signal (0, 1)) obtained in (B) Dot 1 with a value of 1 ”) sum of conformance T ₁ and area 1
The average goodness of fit S ₁ = T ₁ / n is calculated.

In the above example, as shown in FIGS. 8 (b) and 8 (c), the sum T _{1 of the} goodnesses of fit calculated by the coordinate values of n = 35 black areas = 28, the average per area Goodness of fit S ₁
= 28/35 = 0.8.

(D) Calculation of the degree of fitness of the ideal shape As in (C), the figure of the most ideal shape was input.
Sum of goodness of fit T ₂ And the average goodness of fit S₂ Ask for. Most
In case of ideal "8", it is shown in Fig.8 (d) (c).
Sea urchin, the sum T of the fitness of 40 black areas₂ = 38, 1 area
Goodness of fit S per hit₂ = 38/40 = 0.95.

(E) Calculation of certainty factor From the results of (C) and (D), the ratio T ₁ /
The confidence factor is calculated by adding the ratio S ₁ / S ₂ of T ₂ and the average goodness of fit and dividing by 2. In the above example, the certainty factor is

[0073]

[Equation 13]

It becomes

The ratio T ₁ / T ₂ of the sum of the goodness of fit is
In some cases, it exceeds "1", but in that case, it is unconditionally set to "1". That is, min (T ₁ / T ₂ , 1).

The above is the overall evaluation performed by the second main processor 4.

Next, the partial evaluation performed by the first main processor 3 will be described.

The numeral "2" presented in the conventional example,
The case of recognizing “4” will be described. First, FIG. 9 and FIG.
As shown in 0, a fuzzy template for characteristic partial evaluation of the numbers “2” and “4”, that is, a feature extraction fuzzy template is created. It should be noted that when the coordinates of this fuzzy template are constituted by i (i = 1 to 12 dots) and j (j = 1 to 21 dots) in the x and y directions, as shown in FIG. Are numbers "2" and "4"
The fuzzy template for feature extraction can be expressed by the following mathematical formula.

[0079] partial evaluation of the number "2" (feature extraction) for the fuzzy _{template: f = max (f 1,} f 2) ··· (19) However, fuzzy template exists only part of f ≧ 0.

Here,

[0081]

[Equation 14]

[0082] partial evaluation of the number "4" (feature extraction) for the fuzzy _{template: f = max (f 1,} f 2) ··· (22) However, fuzzy template exists only part of f ≧ 0.

Here,

[0084]

[Equation 15]

FIGS. 12 (a) and 12 (b) show examples of the input pattern of the numeral "2" disturbed by noise or the like, and FIG. 13 (a)
(B) shows an example of the input pattern of the number "4".

As a comparative example, a pattern for partial evaluation (feature extraction) is created for the numbers “0”, “6” and “8” as in the case of the numbers “2” and “4”, and FIG. ) (B)
When the certainty factor of recognition was calculated by comparing the input pattern “2” of “1” with these feature extraction patterns, the following results were obtained.

[0087]

[Table 1]

[0088]

[Table 2]

As described above, there is a large difference between the certainty factor when the partial evaluation is performed and the certainty factor when the whole evaluation is performed. Further, by adjusting the coefficient α for weighting the certainty factor when partial evaluation is performed and the certainty factor when overall evaluation is performed, it is possible to construct a system for performing evaluation close to certainty factor of human recognition. it can.

[0090]

As described above, according to the present invention, the operation for calculating the certainty factor of recognition by comparing the entire and characteristic portions of the pattern to be recognized with the overall evaluation pattern and the partial evaluation pattern, respectively. Since it is performed selectively, even when the input pattern is disturbed by noise or the like, the important feature portion of the input pattern can be evaluated. In addition, by calculating the certainty factor of recognition by comparing the entire and characteristic portions of the pattern to be recognized with the overall evaluation pattern and the partial evaluation pattern, respectively, and by weighting this certainty factor, the final recognition certainty factor Can be reflected in.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a fuzzy pattern recognition device according to the present invention.

FIG. 2 is a flowchart showing an operation procedure of the fuzzy pattern recognition device in FIG.

3 is an explanatory diagram showing an example of a three-dimensional membership function used in the fuzzy pattern recognition device in FIG.

FIG. 4 is an explanatory diagram showing a case where a straight line is added to the three-dimensional membership function of FIG.

FIG. 5 is an explanatory diagram showing an example of a character graphic identified in the embodiment and a binary signal corresponding to its shape.

FIG. 6 is an explanatory diagram showing the shape of a three-dimensional membership function used in the embodiment.

FIG. 7 is an explanatory diagram showing the shape of another three-dimensional membership function used in the embodiment.

FIG. 8 is an explanatory diagram showing a procedure for identifying character information in FIG.

FIG. 9 is an explanatory diagram showing a fuzzy template for partial evaluation of the numeral “2”.

FIG. 10 is an explanatory diagram showing a fuzzy template for partial evaluation of the numeral “4”.

FIG. 11 is an explanatory diagram showing coordinates of a fuzzy template.

FIG. 12 is an explanatory diagram showing an input pattern of the numeral “2”.

FIG. 13 is an explanatory diagram showing an input pattern of the numeral “4”.

FIG. 14 is an explanatory diagram showing an example of a characteristic part of numbers.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Image reading device, 2 ... Signal conversion part, 3-5 ... Main processor, 11 ... Storage device, 12 ... Three-dimensional membership function generator, 13 ... Man-machine interface, 14
... signal converter, 15 ... output device.

Claims (3)

[Claims] 1. A fuzzy pattern recognition apparatus for recognizing a pattern based on a membership function representing a fuzzy set, a pattern input section for inputting a pattern to be recognized, and one or more of the membership functions. Collation pattern forming means for forming an overall evaluation pattern and a partial evaluation pattern for recognizing the entire pattern and the characteristic part of the recognition target pattern input through the pattern input unit by combining the functions, And an arithmetic processing unit for selectively performing an operation for calculating the certainty factor of recognition by comparing the whole and characteristic portions of the pattern to be recognized with the overall evaluation pattern and the partial evaluation pattern, respectively. Fuzzy pattern recognition device. 2. The fuzzy pattern recognition apparatus according to claim 1, wherein the partial evaluation pattern can be set only in a region of a characteristic portion of the pattern to be recognized. 3. The calculation processing means weights the certainty factor calculated by comparing the entire and characteristic portions of the pattern to be recognized with the overall evaluation pattern and the partial evaluation pattern, respectively, and finally The fuzzy pattern recognition device according to claim 1, wherein a certainty factor of accurate recognition is calculated.