JPH04263374A - Recognizing and classifying method for optical pattern - Google Patents

Abstract

PURPOSE:To recognize and classify as many patterns to be tested as possible reducing the number of the reference patterns to be recorded by determining the correlation degree between the patterns belonging to some classes to be recognized and classified and the reference patterns. CONSTITUTION:A pattern to be recognized and classified displayed on a picture display device 16 and the reference pattern group are read by coherent luminous flux 12 emitted from a laser 11 and the congruent Fourier conversion pattern is formed on a screen 31 by a Fourier conversion lens 21. This congruent Fourier conversion pattern is read in a two-dimensional photoelectric conversion element 32 displayed on a electric address type liquid crystal light valve 35 by a liquid crystal driving circuit 33 and read by luminous flux 37. The Fourier conversion is again performed for the congruent Fourier conversion pattern by a Fourier conversion lens 41 and the correlation strength on a screen 42 is detected by a two-dimensional photoelectric conversion element 43. Then, the recognition and classification for unknown pattern is performed by comparing with the membership function based on the correlation degree between each pattern to be recognized and each reference pattern.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical pattern recognition classification method used in the field of optical information processing. That is, the present invention relates to an optical pattern recognition and classification method using recognition and associative processing and classification processing, in particular, arithmetic processing of information processing in the field of optical measurement and image processing.

0002

2. Description of the Related Art Conventionally, as methods for performing optical correlation calculations, there have been the matched filter method and the joint transform method. The former method uses a two-dimensional reference pattern.
After Fourier transforming the signal, a reference wave is irradiated and the so-called Fourier transform is applied.
Create a Rie transform hologram and use it as a filter,
By superimposing the Fourier transform image of the test pattern,
It performs correlation calculations. The latter method records a joint Fourier transformed image of the test pattern and reference pattern as an intensity pattern, and performs a correlation calculation by irradiating this with a plane wave. .

However, in both methods, as many reference pattern groups as there are patterns to be identified and classified are prepared.
There is a method in which the pattern that obtained the maximum correlation output in response to the presentation of the test pattern, that is, the result obtained based on the autocorrelation output, is used as the detection result, so it is possible to increase the number of reference patterns. At the same time, the number of reference patterns to be memorized increases, and the burden on the two-dimensional or three-dimensional spatial light modulator increases, and its ability as an identification and classification device is also dominated by the ability of the spatial light modulator. was large and could not be used as a substantially superior optical identification association device.

[0004] The inventors also introduced a feedback system that changes the amount of light irradiated to the reference pattern based on the degree of correlation obtained by the latter method, thereby increasing the number of reference patterns. Although it has been shown that the efficiency of identifying and associating test patterns can be improved by increasing the number of test patterns, it is desired to further increase the number of test patterns that can be substantially recognized and classified. In addition, many recognition classification methods using neural networks have been proposed, but if it becomes necessary to add a new pattern to be recognized and classified after learning, it is necessary to relearn the connection weights and create a new pattern. It was unreasonable that it would take a lot of time to study.

[0005]

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and is capable of identifying a large number of test patterns while greatly reducing the number of reference patterns to be recorded. An object of the present invention is to provide an optical pattern recognition classification method that can perform classification.

[0006]

[Means for Solving the Problems] The present invention has been made in order to solve the above technical problems, and it is an optical pattern.
In the pattern recognition classification method, the correlation output between the test pattern belonging to the class to be identified and classified and multiple reference patterns is obtained multiple times, and a parameter indicating the representative value and dispersion of each obtained correlation degree set is obtained. - Members corresponding to the multiple reference patterns mentioned above for each class that you want to recognize and classify based on the
After creating each ship function, calculate the membership between each correlation output between the test pattern and the plurality of reference patterns and each membership function assigned to each class to be recognized and classified. - Take each membership value and select the smallest membership value of the obtained membership values, or
The optical pattern recognition and classification method is characterized in that the average value of the membership values is determined to be the extent to which the test pattern belongs to the class to be recognized and classified.

[0007]Then, each correlation output with the plurality of reference patterns takes the peak value of the autocorrelation output or cross-correlation output between the test pattern and each of the plurality of reference patterns. And it is suitable. In addition, each correlation output with a plurality of reference patterns is determined based on the total light amount of the light receiving range or It is preferable to use the average light amount. Further, each correlation output with the plurality of reference patterns is obtained by taking the peak value of the autocorrelation output or cross-correlation output between the test pattern and each of the plurality of reference patterns, and It is preferable to set the total amount of light or the average amount of light in the light receiving range according to the spread of the correlation output.

[0008] It is preferable that each correlation output with a plurality of reference patterns is obtained using a matched filter. In addition, each correlation output with a plurality of reference patterns is obtained by individually performing joint Fourier transform of the test pattern and each reference pattern. ,again,
Each intensity pattern obtained by optically performing Fourier transform individually or jointly performing joint Fourier transform of the test pattern and each reference pattern. It is preferable to use a method obtained by optically performing Fourier transform on the -n. Then, each correlation output with a plurality of reference patterns is obtained by superimposing each reference pattern on the test pattern represented by the light transmittance distribution or light reflectance distribution, and then combining it with the incoherent light beam. It is preferable to set the amount of light to be the amount of light emitted and reflected or transmitted.

[0009]

[Operation] According to the optical pattern recognition and classification method as described above, patterns belonging to several classes to be recognized and classified are
By calculating the degree of correlation between the pattern and the reference pattern, the degree of correlation corresponding to the shared features included in each reference pattern can be calculated for each pattern to be recognized and classified by the number of reference patterns. The set of correlation degrees obtained is expressed as a membership function that takes into consideration the amount of fluctuation in the correlation degree due to temporal instability of the spatial light modulator and spec noise.

Next, when identifying and classifying the test pattern, the member is selected based on the degree of correlation with each reference pattern.
By comparing the membership function with the membership function, the degree of matching with the class to be recognized and classified can be obtained from the minimum value or average value of the membership values.

Next, the optical correlation processing method of the present invention will be specifically explained using examples, but the present invention is not limited thereto.

0012

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a schematic diagram showing the functions of an optical correlation processing apparatus according to an example of the optical correlation processing method of the present invention.

In the optical layout diagram of FIG. 1, the optical correlation processing device includes image output means 1, optical Fourier transform means 2, image output means 3, optical Fourier transform means 4, and photodetection means. Since this is a known joint Fourier transform optical system essentially consisting of 5 and 5, its structure will be briefly explained.

One pattern to be recognized and classified drawn on the image display device 16 and a group of reference patterns are read out with the coherent light beam 12 emitted from the laser 11, and the Fourier transform lens 21 converts the pattern into a screen. A congruent Fourier transform pattern is formed on the curve 31. This congruent Fourier transform pattern is read by a two-dimensional photoelectric conversion element 32, drawn on an electrically addressed liquid crystal light valve (hereinafter abbreviated as LCLV) 35 by a liquid crystal drive circuit 33, and read out by a luminous flux 37. The Fourier transform lens 41 performs Fourier transform again, and the correlation strength on the screen 42 is detected by the two-dimensional photoelectric conversion element 43.

In the method of the present invention, after the degree of correlation between the pattern to be recognized and classified and each reference pattern is detected by the optical system, the computer 51 is used to identify each pattern to be recognized. Based on the degree of correlation with each reference pattern, a membership function corresponding to the reference pattern is created using the method described below, and recognition classification of unknown patterns is performed using the membership function. It is characterized by doing things by comparison.

An example of a method for creating the membership function described above will now be described. For example, select two character patterns E and V as a reference pattern group, set the character pattern you want to recognize and classify as H, and arrange the two character patterns on concentric circles as shown in Figure 2. Assume that a character pattern to be recognized and classified is drawn at the center of concentric circles on the image display device 16 at intervals. In this case, the output of the two-dimensional photoelectric conversion element 43, that is, the correlation output between the character pattern H and each reference character pattern appears at each corresponding reference character position drawn on the pattern display device. , as shown in FIG. 3, is detected as light intensity.

[0017] Here, the larger the correlation value, the larger the circle mark. This is because the two parallel lines and horizontal bars that make up H react with the horizontal bars and vertical bars of each reference character pattern, resulting in the difference in correlation output strength as shown in the figure. be. Note that in FIG. 3, the autocorrelation appearing on the optical axis is omitted because it is not relevant here.

Now, the correlation output strength obtained in this way is sent to the computer 51 and stored in memory, and similarly, the patterns to be identified and classified are presented one after another on the image display device 16, and the obtained The peak value of the correlation strength is further calculated by computer 5.
1 one after another. Furthermore, these operations are performed multiple times for each pattern to be identified and classified. This is because the peak value of the correlation output intensity is affected by spec noise caused by coherent light, temporal instability of the image display device 16 and LCLV 35, and input timing of the signal of the two-dimensional photoelectric conversion element 32. This is because the correlation output value fluctuates, so if recognition classification is performed using only one correlation output value, accurate identification classification cannot be performed. Therefore, multiple dates
The average value and standard deviation of the peak values for the patterns belonging to each class are calculated from the data, and the membership function of the peak value of the correlation output strength for the E and V patterns is set to be trapezoidal. For example, 3 times the size of the standard deviation from the average value is set as the membership value 1, 3 times the size of the standard deviation is set as the membership value 0, and the part corresponding to the hypotenuse of the trapezoid is set as the size of the standard deviation. The size is from the same size to 3 times the size.

FIG. 4 is a graph expressing the peak value of the correlation output for E and V as a membership function, where H is the pattern to be identified and classified obtained in this manner. In this way, membership functions are created for all classes to be recognized and classified, and the computer 5
1, along with information on the class that you want to recognize and classify. However, since there are cases where the standard deviation amount happens to be very small in multiple trials, it is necessary to add some correction.

Next, in FIG. 2, an unknown character pattern is presented in place of the character pattern H, and the peak value of the correlation output strength for each reference pattern is calculated using the method described above. , and is input into the computer 51 and compared with the membership functions of the patterns to be recognized and classified. For example, assume that the membership functions for patterns H, W, and N to be recognized and classified are shown in FIGS. 4, 5, and 6, respectively. At this time, it is assumed that the peak value of the correlation output strength of the unknown character pattern is output at point a for pattern E and point b for pattern V. FIG. 7 is a graph showing how much membership value these points a and b have for each pattern to be recognized and classified. In the above example, only the values 1 and 0 were shown, but
In some cases, the part corresponding to the hypotenuse of the trapezoid has a value, and generally the membership value is an analog value.

Now, from the membership values shown in FIG.
If you perform a fuzzy logical AND operation, you will take the minimum membership value of each pattern you want to recognize and classify, and for unknown patterns, patterns H, W, and N are , respectively, become 1, 0, 0. Therefore, it can be said that the unknown pattern is 100% pattern H. On the other hand, the pattern where the minimum membership value is not 0 is
If there are several, it is possible to estimate the extent to which a certain pattern exists based on the minimum value.

Furthermore, the arithmetic mean value of the membership values shown in FIG. 7 can be used as a criterion. That is, in this case, for an unknown pattern, patterns H, W, and H are 1, 0.5, and 0, respectively, so it is possible to estimate the degree of each pattern. can. However, when using the arithmetic mean, an answer may be obtained even if the degree of correlation with a certain reference pattern does not match at all, so the fuzzy logical A
Although the ND operation has fewer errors, it is meaningful as an auxiliary operation when it is difficult to obtain an answer using the AND operation. In other words, if for some reason there is a significant deviation from the state when taking the average value, and the membership value is 0 with only one reference pattern and 1 with other reference patterns, the AND Calculation will not be able to give the correct answer. In this case, the arithmetic mean can be used as the second candidate.

In the above embodiment, the membership function is trapezoidal, but any other shape can be used as long as the membership function has no concavity, and is a so-called convex fuzzy set. It doesn't matter if there is. Now, the correlation output intensity can be obtained in the same way as by taking the total light amount of the light receiving range according to the spread of the correlation output, or the average light amount thereof, in addition to the above-mentioned peak value. In general, when the two patterns are A(x, y) and B(x, y), the correlation output is
It is expressed by the formula shown below. I(x', y') = A(x, y)B*(x'-
x, y'-y)dxdy Here, * indicates the amount of complex conjugation.

From this equation, it can be seen that the correlation output has a spread twice as large as the pattern size. Therefore, the peak
The total amount of light within the area or its average amount of light can be calculated from the square value, taking this spread into consideration. FIG. 8 schematically shows the distribution of the amount of correlated light, and the extent to which the two patterns overlap is shown shifted in the vertical direction of the paper. In this case, for example, pattern E and pattern
As shown in FIG. 8a, the correlated light amount of the patterns H reaches a peak value when they completely overlap, and when the patterns deviate by the size of the patterns as shown in FIG. Since one vertical bar of pattern E overlaps with the vertical bar of pattern E, the amount of correlated light can be detected, although it is smaller than the peak value.

FIG. 8c shows the correlation light intensity distribution of pattern E and pattern H created in this way, where the large circle in the center indicates the peak light intensity, and the small circles above and below indicate the peak light intensity. The circles represent the reactions of the horizontal bars, and the small horizontal circles represent the reactions of the vertical bars. Therefore, it is possible to obtain information in which the characteristics of each pattern, which are not seen in the peak light amount, are better reflected in the total light amount or average light amount.

FIG. 9 shows the reference pattern E and the reference pattern E created based on the average light intensity of the light receiving range according to the spread of the correlation output obtained in the actual experiment using the joint Fourier transform described above. The membership function of the pattern (15 alphabetical patterns) to be discriminated and classified for V is shown. The solid line represents the membership function for pattern E, the broken line represents the membership function for pattern V, and the horizontal axis represents the average light amount of the correlation output. However, there is one that has only one membership function, but this was omitted because that one membership function did not fit within the scale of the horizontal axis.

FIG. 10 shows the results obtained from an experiment using the joint Fourier transform. The figure shows the degree of identification and classification of each pattern derived from the fuzzy logical AND operation results obtained when each of the 15 unknown patterns of the alphabet was presented three times. It is. The alphabet on the left is the input pattern, and the numbers in the first and second parentheses are the reference pattern.
It shows the average light intensity of the correlation measured for E and V. Further, the alphabets on the right side are recognized characters obtained by performing a fuzzy logical AND operation of the membership values shown in FIG. 9, and the numerical value is the result of the AND operation multiplied by 100.

There are several confusing patterns as described above, but if you select the one with the highest degree of confusion, you will have identified all patterns using only two reference patterns. This is not seen in conventional correlation processing using general autocorrelation. It can be seen from Fig. 10 that the set of patterns T, L, I and the set of patterns K, M are particularly confusing, but this is because the characteristics of each pattern are similar. Even if it is as is, it can be used for recognition or classification by selecting those with a high degree of correlation. However, even when this influence is completely removed and used for complete recognition, it can be achieved with very simple processing. That is, a confusing membership function between patterns can be realized by incorporating one pattern that does not overlap. This means that if even one membership value is 0, in order to perform a fuzzy logical AND operation,
This is because they recognize it as a different pattern.

Furthermore, even if the number of classes and patterns to be identified and classified increases, this can be handled easily. This is because it is sufficient to simply create membership functions for the increased number of patterns. In this case, operations such as changing all the weights of all connections connecting neurons through additional learning, as in a normal neural network, are completely unnecessary. In addition, with neural networks, when the number of classes and patterns that need to be identified and classified increases, the learning time becomes extremely long due to the increase in the number of neurons. Such a problem does not arise in the classification method.

[0030] Furthermore, when used for purposes such as association,
By extending the spread of the membership function to a larger extent than that which takes into account fluctuations in the optical system, especially toward the side where the correlation strength is low, it is possible to associate the original pattern from the defect pattern. Can be done. Conversely, when associating a main pattern from a pattern to which extra information has been added, the original pattern can be improved by extending the spread of the membership function toward the side with higher correlation strength. can be associated with.

[0031] It can also be used for purposes such as determining whether there is a defect or not. In this case, it is the opposite of the association case, and the member created by taking into account the fluctuations of the optical system.
If it exceeds the spread of the Ship function, for example, if it exceeds three times the standard deviation due to fluctuations in the optical system, it can be used as a defect.

Now, in the above embodiment, reference pattern 2
However, very roughly speaking, considering the ability of the method of the present invention, the dynamic range of the detected correlation output is, for example, 1:100. , if the non-zero membership value range of one membership function has a spread of 10, then for N reference patterns, 1
This means that a 0N orthogonal area is created. Since it is only necessary to insert the pattern to be identified and classified into these, a large number of different patterns can be identified and classified using a very small number of reference patterns. Furthermore, fuzzy AND
Since it uses arithmetic operations, if it makes decisions that allow for ambiguity, there is no need to assign membership functions to completely orthogonal regions, and the number of patterns that can be identified can be further expanded. .

In addition, both the peak value of the correlation output and the average light intensity of the light receiving area according to the spread of the correlation pattern are calculated as members.
When used as a ship function, two pieces of information with different properties can be obtained from one reference pattern. Therefore, the number of reference patterns can be further reduced, and more reliable recognition classification can be performed from a large amount of information.

In this example, a joint Fourier transformer was used as the optical system for obtaining the correlation output, but this case has the advantage that reference patterns can be easily added and rewritten. There is. Furthermore, to improve the discrimination and classification ability,
It is better not to use a joint Fourier transform correlator. The reason for this is
If the reference pattern group and the test pattern are simultaneously Fourier transformed, the contrast ratio of the multiple interference fringe pattern on the screen 31 in FIG. 1 will decrease as the number of reference patterns increases. In addition, as the obtained correlation output decreases, the dynamic range to which the membership function is assigned decreases, which actually reduces the pattern identification ability. Therefore, a correlation processing method that avoids the above-mentioned decrease in correlation output is more efficient. Next, an embodiment using such a correlation processing method will be described.

[0035]

Embodiment 2 FIG. 11 shows an optical system for regenerating a known matched filter. The matched filter 61 uses a pattern obtained by optically Fourier-transforming a plurality of reference patterns in advance and multiplexing the patterns while changing the irradiation direction of a plane wave for each reference pattern. As shown in FIG.
A beam 12 emitted from the laser 11 is expanded to a suitable beam diameter by a beam expander 13, and illuminates a pattern to be identified and classified drawn on an image display device. Next, the light beam having the complex amplitude distribution of the pattern is subjected to Fourier transformation by the Fourier transformation lens 21, and irradiates the multiple interference fringe pattern drawn on the matched filter. At this time, diffracted light is emitted from a reference pattern having the same spatial frequency spectrum as the pattern to be identified and classified in the direction of the plane wave used when the reference pattern was created.

When this diffracted light is focused onto the screen 31 by the condensing lens 22, one of the diffracted lights becomes a reference pattern.
This becomes a cross-correlation pattern between the pattern and the pattern to be identified and classified, and the other becomes a convolution pattern. Therefore, since the position where the cross-correlation output from each reference pattern appears is known in advance, it is possible to measure the peak value or total light intensity of the correlation output intensity at that position. Hereinafter, the process of similarly measuring the correlation output strength and creating the membership function for other patterns to be identified and classified is the same as in the first embodiment, so the explanation will be omitted.

In this case, since the Fourier components of each reference pattern are not superimposed on the matched filter,
By increasing the number of reference patterns, the contrast ratio of the correlation output decreases to a lesser degree. Therefore, since the dynamic range allocated to the membership function is less likely to be degraded, an extremely large number of patterns can be identified using a small number of reference patterns.

Similar effects can also be achieved using other types of joint Fourier transform devices. For example, the joint Fourier transform of each reference pattern of the joint Fourier transform and the pattern to be recognized and classified is performed individually and simultaneously in parallel using a lens array. This eliminates the problem of the Fourier transform components of each reference pattern overlapping and lowering the contrast ratio of the cross-correlation strength between the reference pattern and the pattern to be recognized and classified. Furthermore,
Unlike a matched filter, reference patterns can be increased or rewritten, so more flexible processing can be performed. For example, in recognition, when complete separation cannot be performed with a preset reference pattern, it is possible to respond to a request to rewrite part of the reference pattern.

Both embodiments 1 and 2 are characterized in that they are resistant to translational deviations of input patterns because they are based on coherent optical diameters. This is because the correlation output position moves in response to the translational shift of the input pattern.

Next, an example of an optical system that is simple and relatively easy to integrate will be described.

Embodiment 3 The example shown in FIG. 12 is a known incoherent correlation optical system. As an optical system, the image display device 16 and the reference pattern mask 62 are arranged close to each other, and the output of the product of the pattern drawn on the image display device 16 and the reference pattern is A condensing lens array 71 is used to condense light onto the screen 31, and a two-dimensional photoelectric conversion element 43 detects the output from each reference pattern condensed. On the image display device 16, patterns to be identified and classified are displayed in an array for pixels corresponding to the reference pattern mask 62.

In this case, the cross-correlation peak between each reference pattern and the pattern to be identified and classified will be detected. The following operation displays other patterns to be identified and classified one after another on the image display device, and based on these output values,
The same effect as in the first and second embodiments can be achieved by creating as many membership functions as there are reference patterns for each pattern to be identified and classified. However, there is no tolerance for translational deviation of the input pattern,
It has a very simple structure and is easy to integrate.

Note that any type of optical correlation processing system may be used in the above embodiment. Furthermore, various optical systems have been proposed for the joint Fourier transform correlation optical system, but the screen 31 and the two-dimensional photoelectric conversion element 3 shown in FIG.
2. The same effect can be obtained by using an optically addressed liquid crystal light valve instead of the pattern processing device 33 and the electrically addressed liquid crystal light valve 35.

[0043] In the present invention, although there are differences in specifications regarding the part that functions as a spatial light modulator, in principle they are all the same electrically addressed type and optically addressed type. is available. Furthermore, analog and digital types can also be used. Most of the devices described below belong to analog devices, but devices using ferroelectric materials correspond to digital devices. In addition to the above-mentioned liquid crystal light valves, examples of electrically addressed types include PLZT, KDP,
Those with an electro-optic effect, such as BSO, are often used.

[0044] Even in the case of the photo-addressable type, a photoconductive layer is generally combined with the same material as the electrically addressable type. However, in crystals having a photovoltaic effect such as BSO and BaTiO3, a photo-induced refractive index change occurs due to spontaneous polarization depending on the input light intensity, so there is no need to add a photoconductive layer. Note that these spatial light modulators can be configured as either a transmissive type or a reflective type.

However, if the read light completely cancels out the information of the write light in the optical addressing type, the wavelength ranges of the read light and the write light may be separated so that the read light does not affect the write information. It is necessary to take measures such as not giving Further, when using an electrically addressed type, a two-dimensional photoelectric conversion element and a driving circuit therefor are required to obtain the input information, but there is an advantage that the signal can be easily processed.

[0046]

Effects of the Invention As explained above, the above-mentioned effects were obtained by the optical pattern recognition classification method of the present invention. Putting these together, we get the following remarkable technical effects. That is, first, by presenting a small number of reference patterns, it was possible to provide an optical pattern recognition method that can recognize and classify a large number of patterns.

Second, by including temporal fluctuations of the spatial light modulator and time-series fluctuations such as speckle noise in the membership function of the degree of correlation created based on various statistical quantities. We were able to provide an optical pattern recognition classification method that can eliminate the uncertainty of judgment in analog processing.

Third, by performing a fuzzy logical AND operation of the membership functions of the correlation degrees of each reference pattern that the pattern to be recognized and classified has, an optical Therefore, we were able to provide a pattern recognition classification method. Fourthly, even if a new pattern is added to be recognized and classified, it can be easily executed by creating a membership function of the degree of correlation of each reference pattern to this pattern. An optical pattern recognition classification method could be provided.

Fifth, even if the recognition result is confusing,
To provide an optical pattern recognition classification method that can easily and completely recognize a pattern by adding a small number of reference patterns and creating a new membership function for this pattern. was completed. Sixth, recognition, classification,
By adjusting the spread of the membership function separately from the fluctuation of the optical system depending on the purpose of association, etc., we were able to provide an optical pattern recognition method that allows flexible processing. .

[Brief explanation of the drawing]

FIG. 1 shows an example of a joint frame used as an optical correlation processing device in the optical pattern recognition classification method of the present invention.
FIG. 2 is a schematic configuration diagram showing a Rie transform optical system.

[Fig. 2] In the optical pattern recognition and classification method of the present invention, one pattern to be recognized and classified and a group of reference patterns when a joint Fourier transform optical system is used as the optical correlation processing device. FIG.

FIG. 3 is a diagram showing the state of correlation output when a joint Fourier transform optical system is used as an optical correlation processing device in the optical pattern recognition classification method of the present invention.

FIG. 4: In the optical pattern recognition and classification method of the present invention, each reference pattern for one pattern to be recognized and classified
FIG. 3 is a diagram showing a membership function created based on the degree of correlation between members.

FIG. 5 is a diagram showing membership functions created based on the degree of correlation of each reference pattern with another pattern to be recognized and classified in the optical pattern recognition and classification method of the present invention. be.

FIG. 6 shows a membership function created based on the degree of correlation of each reference pattern with another pattern to be recognized and classified in the optical pattern recognition and classification method of the present invention. It is a diagram.

[Fig. 7] In the optical pattern recognition and classification method of the present invention, the degree of correlation measured for an unknown pattern represents the membership value that it has for the pattern to be recognized and classified. This is a diagram.

FIG. 8 is a diagram schematically representing a cross-correlation pattern used in the optical pattern recognition classification method of the present invention.

FIG. 9 is a diagram showing the membership values of 15 patterns to be recognized and classified for two experimentally obtained reference patterns in the optical pattern recognition classification method of the present invention.

FIG. 10 is a diagram showing experimentally obtained recognition results for unknown patterns using fuzzy AND operations and the accuracy of the recognition results in the optical pattern recognition classification method of the present invention. .

FIG. 11 is a schematic configuration diagram of an optical system using a matched filter as another example used as an optical correlation processing device in the optical pattern recognition classification method of the present invention.

FIG. 12 is a schematic configuration diagram of yet another example of an optical system used as an optical correlation processing device in the optical pattern recognition classification method of the present invention.

[Explanation of symbols]

1, 3 Pattern output means 2, 4
Optical Fourier transform means 5
Light detection means 11
Laser 12, 37 Luminous flux 13 Beam expander 4,
22, 39 Beam splitter 35
LCD light bulb 16
Pattern display devices 21, 41 Fourier transform lenses 31, 42 Screens 32, 43 Two-dimensional photoelectric conversion elements 33, 51
Pattern processing devices 34, 38
mirror

Claims (7)

[Claims] Claim 1: In an optical pattern recognition classification method, correlation outputs between a test pattern belonging to a class to be classified and a plurality of reference patterns are obtained multiple times, and each obtained correlation degree is calculated. After creating membership functions corresponding to the plurality of reference patterns described above for each class to be recognized and classified based on the representative value of the set and the parameter indicating the dispersion, the test pattern is and the plurality of reference patterns, and the membership value of each membership function assigned to each class to be recognized and classified, and the obtained membership value. The optical pattern is characterized in that the smallest membership value of the optical pattern or the average value of the membership values is set to the extent that the test pattern belongs to the class to be recognized and classified. Recognition classification method. 2. Each correlation output with the plurality of reference patterns takes a peak value of an autocorrelation output or a cross-correlation output between the test pattern and each of the plurality of reference patterns. 2. The optical pattern recognition classification method according to claim 1. 3. Each correlation output with the plurality of reference patterns has a light receiving range according to the spread of the autocorrelation output or cross-correlation output between the test pattern and each of the plurality of reference patterns. Claim 1 characterized in that it is the total light amount or the average light amount of
The optical pattern recognition classification method described in . 4. Each correlation output with the plurality of reference patterns is a peak value of an autocorrelation output or cross-correlation output between the test pattern and each of the plurality of reference patterns; The optical pattern according to claim 1, characterized in that the total light amount or the average light amount in the light receiving range is determined according to the spread of their correlation outputs.
recognition classification method. 5. The optical pattern recognition and classification method according to claim 1, wherein each correlation output with the plurality of reference patterns is obtained using a matched filter. 6. Each correlation output with the plurality of reference patterns is obtained by individually performing a joint Fourier transform of the test pattern and each reference pattern. putter
The obtained pattern can be optically Fourier-transformed individually again, or joint Fourier-transformation of the test pattern and each reference pattern can be performed all at once. The optical pattern according to any one of claims 1 to 4, wherein the optical pattern is obtained by optically performing Fourier transform on each of the intensity patterns at once. recognition classification method. 7. Each correlation output with the plurality of reference patterns is obtained by superimposing each reference pattern on a test pattern represented by a light transmittance distribution or a light reflectance distribution, and 2. The optical pattern recognition and classification method according to claim 1, wherein incoherent light is irradiated and the amount of reflected or transmitted light is determined.