JP3272506B2 - Optical pattern recognition and classification device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for extracting a specific object or character from an image containing many objects or characters, such as landscapes, and extracting the extracted object, character, image, etc. The present invention relates to an optical pattern recognition / classification apparatus used for pattern recognition and search, inference, association, and the like.

[0002]

2. Description of the Related Art A specific pattern to be extracted is extracted from an image (hereinafter, referred to as an input image) containing a pattern of a large number of objects and characters such as everyday scenery seen by a human, and is recognized and classified. As one of the methods, a matched filtering method is known.

In the matched filtering method, a Fourier transform image of an input image is formed on a back focal plane of a Fourier transform lens, is incident on a matched filter of a pattern to be extracted, which is disposed at a focal position, and transmits transmitted light. When the Fourier transform is further performed by the Fourier transform lens, when the same pattern as the pattern to be extracted exists in the input image,
This is a method in which a correlation spot showing a high correlation appears on the back focal plane of the Fourier transform lens, and the position of the same pattern as the pattern to be extracted in the input image can be known from the position, and recognition and classification can be performed. At this time, a hologram in which the Fourier transform of the intensity distribution of the pattern to be extracted is recorded in the form of an interference fringe with the reference light is generally used for the matched filter.

However, in this method, when a hologram is used as a filter, it is necessary to make a filter for each pattern to be extracted, and the tolerance for the alignment of the filter is extremely small.
There is a problem that the misalignment significantly lowers the performance.

Also, without using a hologram as a filter,
There is also a method using a spatial light modulator that can change the pattern to be extracted in real time. There is a problem that it is inferior.

As another method, there is a template matching method used in a digital image processing device of a computer or the like. This template matching method is a method of moving a pattern to be extracted in an input image and searching for the most suitable place. This method includes a residual sequential test method, a method using a cross-correlation coefficient, and the like. However, any of these methods basically has a problem that a pattern to be extracted must be scanned in an input image, which takes time. This is a problem that appears more noticeably as the number of pixels increases.

Further, in any of the above methods, a pattern present in the input image and a pattern present from outside the input image are provided.
If the pattern to be extracted does not substantially match, in other words, if the degree of correlation is not very high, it is difficult to extract from the input image, and even if there is only a slight difference in the shape of the pattern, for each of those patterns, for example, The matched filtering method has a disadvantage that a plurality of filters must be manufactured.

In order to make the pattern extraction more flexible, there has been proposed a method of extracting a pattern from an input image using a special filter called SDF or MACE. In this case, too, it is desired to extract the pattern in advance. There is a problem that a complicated filter corresponding to the pattern must be designed and manufactured.

[0009]

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems in the prior art.
Provided is an optical pattern recognition / classification apparatus which does not use a matched filter which draws interference fringes such as holograms which must be produced for each pattern to be extracted, and which relatively easily aligns an optical system. It is intended for that purpose. In addition, the present invention is not limited to the scanning of the pattern to be extracted, and from the image including the pattern of many objects and characters, not only the pattern that is the same as the pattern to be extracted, but also a certain degree of correlation. It is an object of the present invention to provide an optical pattern recognition / classification device that can easily and simultaneously extract patterns by using a simple filter and perform finer recognition / classification for each extracted pattern.

[0010]

SUMMARY OF THE INVENTION The present invention has been made to solve the above technical problems. An optical pattern recognition / classification device (1) of the present invention converts a pattern belonging to a desired large category including a set of one or more small categories including one or more patterns to be recognized / classified into an input image (hereinafter referred to as an input image). A pattern extracting unit that extracts from among the images (hereinafter, referred to as an image) and a pattern (hereinafter, referred to as an extracted pattern) extracted by the pattern extracting unit (hereinafter, referred to as an extracted pattern) are correlated or similar to each other. A feature amount detecting means for detecting using an optical system for performing an operation;
Discriminating means for discriminating the small category to which the extracted pattern belongs, based on one or more feature amounts,
A first light source that emits a first coherent light beam; and a first image display that gives an intensity distribution to the first coherent light beam by displaying an input image as an intensity transmittance or a reflectance distribution. Means, a first Fourier transform lens for optically Fourier transforming the intensity distribution of the light beam transmitted or reflected by the first image display means, and a feature common to the entire pattern included in the desired large category. First Fourier transform image output means for outputting a Fourier transform image of the pattern (hereinafter, referred to as a representative pattern), and a Fourier transform image of the representative pattern, which is arranged on the Fourier transform surface of the first Fourier transform lens. ,
The intensity transmittance or reflectance distribution is displayed and the first
A first spatial light modulator that further provides an intensity distribution to the luminous flux of
A second Fourier transform lens for optically further Fourier transforming the intensity distribution of the light beam transmitted or reflected by the first spatial light modulator, and an output image obtained by Fourier transform by the second Fourier transform lens Imaging means for imaging the image, image output means for displaying the output image as an input image of the first image display means again, and detecting a pixel having an output value equal to or more than a certain value in the output image,
The position detection means detects the position of a group of a certain number or more of the pixels, and the area cutout means for cutting out an area of an appropriate size from the input image based on the output from the position detection means. And the image area cut out by the area cutout means corresponds to the pattern extracted by the pattern extraction means, and the feature amount detection means emits a second coherent light beam And second image display means for displaying one or a plurality of reference patterns as intensity transmittance or reflectance distribution to give an intensity distribution to the second coherent light flux, and second image display means Third optically Fourier transforms the intensity distribution of the light beam transmitted or reflected by
A Fourier transform lens for outputting a Fourier transform image of the extracted pattern obtained by the pattern extracting means.
A Fourier-transformed image output means, and a Fourier-transformed image of the extracted pattern, which is arranged on a Fourier-transformed surface of the third Fourier-transformed lens, is displayed as an intensity transmittance or a reflectance distribution, and the second A second spatial light modulator that further provides an intensity distribution to the light beam, a fourth Fourier transform lens that optically further Fourier-transforms the intensity distribution of the light beam transmitted or reflected by the second spatial light modulator, Light receiving means for detecting the amount of light of the image obtained by Fourier transform by the fourth Fourier transform lens in accordance with the one or more reference patterns, and detecting the light quantity by the light receiving means The amount of light of the image corresponds to the feature amount detected by the feature amount detection means.

Further, another optical pattern recognition / classification apparatus (2) of the present invention includes a pattern belonging to a desired large category consisting of a set of one or more small categories including one or more patterns to be recognized and classified. A pattern extraction unit that extracts from an input image (hereinafter, referred to as an input image), and an amount (hereinafter, referred to as a feature amount) representing each characteristic of a pattern (hereinafter, referred to as an extraction pattern) extracted by the pattern extraction unit ) Using an optical system that performs correlation or similar operation,
Discriminating means for discriminating the small category to which the extracted pattern belongs based on one or more feature amounts detected by the feature amount detecting means, wherein the pattern extracting means comprises a first coherent light beam. A first image display means for displaying an input image as an intensity transmittance or a reflectance distribution to give an intensity distribution to the first coherent light beam, and a light source for transmitting the first image display means. Alternatively, a first Fourier transform lens for optically Fourier transforming the intensity distribution of the reflected light beam, and a Fourier transform of a pattern (hereinafter, referred to as a representative pattern) having characteristics common to all the patterns included in the desired large category. First Fourier transform image output means for outputting an image;
A first space that is arranged on the Fourier transform surface of the Fourier transform lens of the above and displays the Fourier transform image of the representative pattern as an intensity transmittance or a reflectance distribution to further provide an intensity distribution to the first light flux. An optical modulator; a second Fourier transform lens for optically further Fourier transforming the intensity distribution of the light beam transmitted or reflected by the first spatial light modulator; and a Fourier transform by the second Fourier transform lens. Imaging means for capturing the obtained output image, image feedback means for displaying the output image again as an input image of the first image display means, and having an output value equal to or more than a certain value in the output image A position detecting means for detecting a pixel and detecting a position of a group of a certain number or more of the pixels; and A region extracting unit for extracting a region, and the image region extracted by the region extracting unit corresponds to a pattern extracted by the pattern extracting unit; A light source that emits a coherent light beam, a second image display unit that displays one or a plurality of reference patterns with an intensity transmittance or a reflectance distribution, and gives an intensity distribution to the second coherent light beam, Each lens is arranged corresponding to the position of the one or a plurality of reference patterns, and a first element for optically Fourier-transforming the intensity distribution of a light beam transmitted or reflected by the second image display means for each reference pattern. A Fourier transform lens array,
A second Fourier transform image output unit that outputs a Fourier transform image of the extracted pattern obtained by the pattern extracting unit, and an image of the Fourier transform image of the extracted pattern.
An image processing device that is arranged in parallel at a position corresponding to each lens of the first Fourier transform lens array; and an image processing device that is arranged on the Fourier transform surface of each lens of the first Fourier transform lens array and created by the image processing device A third spatial light modulator that displays the parallel image of the Fourier transform image of the extracted pattern as an intensity transmittance or a reflectance distribution to further provide an intensity distribution to the second light flux, A second Fourier transform lens array arranged corresponding to the position of one or a plurality of reference patterns and optically further Fourier-transforming the intensity distribution of the light flux transmitted or reflected by the third spatial light modulator; The amount of light of the third output image obtained by Fourier transform by each lens of the second Fourier transform lens array is compared with the one or more reference patterns. Amount of image to be detected by being composed of a light receiving means, and said light receiving means for detecting in the, to correspond to the feature quantity detected by the feature quantity detecting means, characterized in.

Further, in the apparatus according to the present invention, at least one of the first or second Fourier transform image output means is provided.
First, a light source that emits a coherent light beam, and a third image display unit that displays the representative pattern or the extracted pattern as an intensity transmittance or a reflectance distribution to give an intensity distribution to the light beam, A fifth Fourier transform lens for optically Fourier transforming the intensity distribution of the light flux transmitted or reflected from the third image display means, and the representative pattern obtained by the fifth Fourier transform lens;
Alternatively, it is preferable that the imaging device is configured to include an imaging unit that captures a Fourier transform image of the extracted pattern.

In the apparatus of the present invention, it is preferable that at least one of the first, second, and third spatial light modulators is a first optical writing type spatial light modulator. .

Further, in the apparatus of the present invention, the first image display means includes a second optically-written spatial light modulator, and an input image on a writing surface of the second optically-written spatial light modulator. And an imaging lens system for forming an image.

Further, in the above-mentioned apparatus according to the present invention, at least one of the first, second and third spatial light modulators is provided.
Finally, it is preferable that the Fourier transform image of the representative pattern or the extracted pattern is displayed as an image binarized by an appropriate threshold substantially in real time.

Further, in the apparatus according to the present invention, at least one of the first and second optical writing type spatial light modulators displays an image written from a writing surface by binarizing the image with a threshold value. Is preferably performed.

Further, in the apparatus according to the present invention, the discriminating means acquires, for each small category, a plurality of feature amounts of an extracted pattern belonging to the small category, and determines a representative value of a set of obtained outputs as a reference. Then, a membership function of the extracted pattern is created in advance for each component (hereinafter, referred to as a feature parameter) of the extracted pattern and the feature amount, and each small category is extracted from the feature amount of the extracted pattern extracted from the input pattern. A membership value is calculated for each feature parameter based on each of the membership functions described above, and an average value or a minimum value of the obtained membership values is set to a degree belonging to each of the small categories of the extraction pattern. From this degree, a device composed of arithmetic means for determining an extraction pattern is preferably used.

Further, in the apparatus according to the present invention, the discriminating means is a neural network which inputs each characteristic amount acquired for each of the extracted patterns and has an output indicating a small category of the extracted pattern. It is preferable that the determination of the extraction pattern is performed based on.

[0019]

According to the configuration of the optical pattern recognition / classification apparatus of the present invention as described above, the first coherent light beam emitted from the light source is transmitted to the first focal point plane of the first Fourier transform lens. The light enters the image display means. In the first image display means, the input image is displayed with the intensity transmittance or reflectance distribution, so that the incident first coherent light flux is modulated into the intensity distribution of the input image and is transmitted or reflected as light. Emit the first image display means. The modulated first coherent light beam passes through a first Fourier transform lens and forms a Fourier transform image of the input image on a rear focal plane.

On the other hand, a Fourier-transformed image of a pattern (hereinafter, referred to as a representative pattern) having characteristics common to the entire pattern belonging to a desired large category consisting of a set of one or more small categories including one or more patterns is obtained. The image is output by one Fourier transform image output means, and is displayed on the first spatial light modulator as an intensity transmittance or reflectance distribution. Since the first spatial light modulator is arranged on the rear focal plane of the first Fourier transform lens, the Fourier transform image of the input image is incident on the first spatial light modulator, and the Fourier transform of the representative pattern is performed. The output is further modulated by the intensity distribution of the image, and the output of the product of the Fourier transform image of the input pattern and the Fourier transform image of the representative pattern is transmitted or reflected.

As described above, since only the Fourier transform image is drawn on the first spatial modulator and no interference fringes are drawn, a hologram or the like for drawing interference fringes used in a method such as a matched filter method is used. There is no need to use. Therefore, the arrangement of the first spatial light modulator does not need to be stricter than that of the matched filter. Moreover, when changing the representative pattern, it is only necessary to change the Fourier transform image of the representative pattern drawn on the first spatial light modulator, and this can be easily performed by using a computer or the like. Furthermore, the representative pattern may be a simple one such as a circle or a square, and does not require a complicated one.

Thereafter, the first coherent light beam is again Fourier-transformed by the second Fourier transform lens disposed after the spatial light modulator by its own rear focal length,
An output image is obtained at the front focal position of the second Fourier transform lens. This output image is a result of the cross-correlation operation between the input image and the representative pattern, and it can be said that a portion having a spatial frequency component similar to the representative pattern remains in the input image.

The output image is picked up by an image pickup means.
The captured output image is displayed again on the first image display means as an input image by the image feedback means. The series of processes described above is repeated again. If the number of repetitions is appropriately controlled, only the pattern that is correlated with the representative pattern in the input image is output to the end, and after all, the pattern belonging to the desired large category is extracted from the input image. The obtained output image can be obtained. In addition, the above-described processing performs the entire input image simultaneously and in parallel with one coherent light beam, so that there is no need for scanning time such as template matching.

The output image finally picked up is input to the position detecting means. The position detection means detects a pixel having an output value equal to or higher than a certain level in the output image, and detects a position of a group in which a certain number or more of the pixels are collected. Based on this position information, an area of an appropriate size corresponding to the size of the group of pixels is cut out from the original input image, and an extraction pattern is obtained. With this configuration, one or more patterns belonging to a large category formed by a group of one or more small categories are extracted from an input image including patterns of many shapes of objects and characters. Therefore, not only patterns having high correlation with the representative pattern but also patterns having some correlation can be simultaneously extracted.

Further, the second coherent light beam emitted from the light source enters the second image display means disposed at the front focal position of the third Fourier transform lens. In the second image display means, one or a plurality of reference patterns are displayed with an intensity transmittance or a reflectance distribution, so that the incident second coherent light beam has an intensity distribution of the one or a plurality of reference patterns. And is emitted from the second image display means as transmitted or reflected light. The modulated second coherent light beam passes through a third Fourier transform lens and forms a Fourier transform image of one or more reference patterns on the rear focal plane.

On the other hand, a Fourier-transformed image of the extracted pattern extracted from the input image is output by the second Fourier-transformed image output means, and is displayed on the second spatial light modulator in the form of intensity transmittance or reflectance distribution. . Since the second spatial light modulator is arranged on the focal plane of the third Fourier transform lens, a Fourier transform image of one or a plurality of reference patterns is incident on the second spatial light modulator, and Is further modulated by the intensity distribution of the Fourier-transformed image, and the output of the product of the Fourier-transformed image of one or more reference patterns and the Fourier-transformed image of the extracted pattern is transmitted or reflected.

As described above, since only the Fourier transform image is drawn on the second spatial light modulator and no interference fringes are drawn, a hologram for drawing interference fringes used in a method such as a matched filter method is used. There is no need to use such as. Therefore, the arrangement of the second spatial light modulator does not need to be stricter than that of the matched filter.

Thereafter, the second coherent light beam is again Fourier-transformed by the fourth Fourier transform lens disposed behind the second spatial light modulator by its own rear focal length, and the obtained image is transformed. The light amount is detected by light receiving means arranged corresponding to one or a plurality of reference patterns. The detected output increases as the degree of coincidence between the Fourier transform image of the reference pattern and the Fourier transform image of the extracted pattern on the second spatial light modulator increases, so that the cross-correlation calculation between the reference pattern and the extracted pattern is performed. Can be said to be the output of the result, and the acquisition of each component of the plurality of feature amounts of the extraction pattern can be performed simultaneously in parallel with a simple optical system at high speed. Also, one or more reference patterns and the extracted pattern are compared on the Fourier transform plane,
Even if the extraction pattern moves, output position invariance is ensured.

The feature value detected by the above-described feature value detecting means is input to the discriminating means, and it is determined which of the small categories included in the large category the extracted pattern belongs to.

Further, the second coherent light beam emitted from the light source is incident on the second image display means disposed at the front focal position of each lens of the first Fourier transform lens array. In the second image display means, one or a plurality of reference patterns are displayed with a strong transmittance or reflectance distribution.
Modulated into one or more reference pattern intensity distributions,
The second image display means is emitted as transmitted or reflected light.
The modulated second coherent light beam passes through a first Fourier transform lens array in which each lens is arranged corresponding to the position of one or more reference patterns, and one at a rear focal position of each lens. Alternatively, Fourier transform images of a plurality of reference patterns are formed for each lens.

On the other hand, a Fourier-transformed image of the extracted pattern extracted from the input image is output by the second Fourier-transformed image output means. An image in which a plurality of Fourier transform images of the extracted pattern are arranged in parallel is created. Thereafter, the parallel image of the Fourier transform image of the extracted pattern is displayed on the third spatial light modulator with a strong transmittance or reflectance distribution at a position corresponding to each lens of the first Fourier transform lens array. Since the third spatial light modulator is disposed on the rear focal plane of each lens of the first Fourier transform lens array, the Fourier transform image of one or a plurality of reference patterns is converted to the third spatial light modulator. , And further modulated by the intensity distribution of the Fourier transformed image of the extracted pattern, and the output of the product of the Fourier transformed image of one or more reference patterns and the Fourier transformed image of the extracted pattern is transmitted or reflected.

As described above, since only the Fourier transform image is drawn on the third spatial modulator and no interference fringes are drawn, a hologram or the like for drawing interference fringes used in a method such as a matched filter method is used. There is no need to use. Therefore, the arrangement of the third spatial light modulator is not required to be stricter than that of the matched filter.

Thereafter, each lens of the second Fourier transform lens array is disposed behind the third spatial light modulator by its own back focal length, which causes the second coherent light beam to again be Fourier transformed. The light amount of the converted image obtained at the front focal position of each lens of the second Fourier transform lens array is detected by light receiving means arranged corresponding to one or a plurality of reference patterns. The detected output increases as the degree of coincidence between the Fourier transform image of the reference pattern and the Fourier transform image of the extracted pattern on the third spatial light modulator increases, so that the cross-correlation operation between the reference pattern and the extracted pattern is performed. Can be said to be the output of the result, and the acquisition of each component of the plurality of feature amounts of the extraction pattern can be performed simultaneously in parallel with a simple optical system at high speed. Further, since one or a plurality of reference patterns and the extracted pattern are compared on the Fourier transform plane, even if the extracted pattern moves, the position invariance of the output is ensured.

The feature value detected by the above-described feature value detection means is input to the discrimination means to determine which small category included in the large category belongs to the extracted pattern.

In one or more or all of the first or second Fourier-transformed image output means, the coherent light beam emitted from the light source has a representative pattern or an extracted pattern having an intense transmittance or The light is displayed in a reflectance distribution, and is incident on a third image display means disposed on the front focal plane of the fifth Fourier transform lens.
The incident coherent light beam is modulated into the intensity distribution of the representative pattern or the extracted pattern and emitted as transmitted or reflected light. A Fourier transform image of the representative pattern or the extracted pattern is formed, and the pattern is imaged by the imaging means. By employing such means, it is possible to output a Fourier transform image of the representative pattern or the extracted pattern easily and optically at a high speed simply by displaying the representative pattern or the extracted pattern. The change of the representative pattern or the extraction pattern can be performed only by changing the image displayed on the third image processing means.

Further, by using an optical writing type for at least one or all of the first, second, or third spatial light modulators, the Fourier transform image output means can be used as a representative pattern, Alternatively, the Fourier transform image of the extracted pattern can be directly written in the spatial light modulator, and the Fourier transform image of the representative pattern or the extracted pattern can be displayed optically and quickly and easily.

In the first image display means, an input image is formed directly on a writing surface of a light writing type spatial light modulator by using an image forming lens system, so that an electric means such as a liquid crystal television is provided. The input image can be displayed optically at high speed and easily with the intensity transmittance or the reflectance distribution without using the image, and the image feedback means can also directly output the output image on the writing surface of the light-writing spatial light modulator. By forming an image, high-speed and simple realization can be achieved.

Still further, one or more or all of the first, second, or third spatial light modulators may be provided with a Fourier transform image of the representative pattern or the extracted pattern substantially in real time. By displaying the image as a binarized image using an appropriate threshold value, it is possible to improve the prevention of a decrease in the contrast of the extracted pattern in the output image and the reference pattern incident on the light receiving unit.

In addition, an operation is performed in which one or all of the first or second optical writing type spatial light modulators are displayed by binarizing an image written from a writing surface with a threshold value. With this configuration, it is also possible to improve the decrease in the contrast of the extracted pattern in the output image or the reference pattern incident on the light receiving unit.

In the discriminating means used in the apparatus of the present invention, by creating a membership function for each feature parameter for each extracted pattern, the degree to which the member belongs in the feature space having the dimension of the number of feature parameters is determined. It is possible to create a small category for each extraction pattern represented by a ship function. Further, by acquiring the characteristic parameter a plurality of times and using the representative value as a reference for creating the membership function,
A small category can be created in consideration of output fluctuation. After the creation, the membership value of each component is obtained from the feature amount obtained for each unknown extraction pattern, and the minimum or average value of the obtained membership values is obtained. The degree to which a category belongs can be determined, and the extraction pattern can be determined.

Further, a neural network is used as the discriminating means, and each feature amount obtained by the feature quantity detecting means is input to each node of the input layer of the neural network. , The small category to which the extracted pattern belongs can be easily and effectively recognized and classified. In the neural network, by learning, the weighting of the criterion of the feature parameter having a large contribution to the determination of the pattern is increased, and the weighting of the criterion of the feature parameter having a small contribution is reduced.
When there are a plurality of patterns having similar features, a portion having a difference between those features is particularly emphasized for determination, so that a determination criterion can be easily created.

Next, the optical pattern recognition / classification apparatus of the present invention will be specifically described with reference to examples, but the present invention is not limited thereto.

Embodiment 1 FIG. 1 is a schematic structural view of an optical pattern recognition / classification apparatus showing one embodiment of the present invention. In this embodiment, a pattern belonging to a large category including some small categories is extracted from an input image by a pattern extracting unit 1, and an extraction belonging to the extracted large category is extracted by an optical correlation calculating unit 2 and a discriminating unit 3. A detailed classification of which small category the pattern belongs to is performed.

The coherent light beam 12 emitted from the laser 11 is expanded to an appropriate light beam diameter by a beam expander 13 and collimated, and then enters a half mirror 14 to be split into two light beams. The first light flux 12a enters the first image display means 15. First image display means 15 arranged on the front focal plane of first Fourier transform lens 16
, An input image from which a pattern belonging to a large category is to be extracted is drawn in a transmittance distribution of an intensity pattern,
The input image is read through the light flux 12a. The first light flux 12a transmits through the first Fourier transform lens 16, and forms a Fourier transform image of the input image on the first spatial light modulator 17 arranged on the rear focal plane of the first Fourier transform lens 16.

On the other hand, a pattern (hereinafter referred to as a representative pattern) having characteristics common to a desired large category to be extracted from the input image is stored in the computer 18 in advance, and the first Fourier transform image output means 19 stores the representative pattern. Is output, and the first spatial light modulator 1
In FIG. 7, the intensity pattern of the Fourier transform image of the target pattern is displayed as a transmittance distribution. At this time, the computer 18
In the above, the Fourier transform image of the representative pattern may be stored in advance. Therefore, when the Fourier transform image of the input image is transmitted through the first spatial light modulator 17, it is modulated by the intensity pattern of the Fourier transform image of the representative pattern,
The calculation of the product of the corresponding spatial frequency components of the input image and the representative pattern is performed. Thereafter, the first light flux 12a is transmitted through the second Fourier transform lens 20 and subjected to Fourier transform again. The cross-correlation between the input image and the representative pattern is displayed on a screen 21 arranged on the front focal plane of the second Fourier transform lens 20. The result of the operation is output.

More specifically, the output image is an image in which an input image having a small correlation with the representative pattern is deleted.
The part of the input image that has a high degree of correlation with the representative pattern is
The part that is output without a decrease in contrast and has little correlation with the representative pattern is output with little contrast, and an image in which a pattern similar to the representative pattern remains is output and imaged. An image is taken by the means 22. After the output image is picked up by the image pick-up means 22, it is displayed again on the first image display means 15 as an input image, and the above-described series of processing is repeated an appropriate number of times.
A portion having a low correlation with the representative pattern in the input image disappears, and an image in which only a pattern having a higher similarity with the representative pattern remains until the end is output. At this time, the number of repetitions is controlled to an appropriate number by a program of the computer 18.

Finally, the output image is taken into the computer 18 via the image pickup device 22, and the position of the pattern remaining in the output image is detected by the image processing device 23. A pixel having an output value equal to or more than a certain value in the output image is detected, a group of the pixels having a certain number or more is detected, and information on the position is output. Based on this position information, a region of an appropriate size corresponding to the size of the group of pixels is cut out from the input image to form an extraction pattern. The image processing device 23 can be realized using a general image processing board or the like used by being connected to a personal computer or the like.

According to the above-described processing, one or a plurality of patterns correlated with the representative pattern are extracted and become an extracted pattern. That is, all patterns belonging to the large category having common features have been extracted from the input image as extracted patterns.

In contrast to the above-described apparatus of the present invention, in the conventional apparatus used in the matched filtering method, as shown in FIG. 2, the Fourier transform image of the input image is converted by the first Fourier transform lens 16. The light enters the matched filter 41 for pattern to be extracted, which is formed on the back focal plane and is arranged at this focal position, and the transmitted light is further Fourier transformed by the second Fourier transform lens 20. At this time, if the same pattern as the pattern to be extracted exists in the input image, a correlation spot showing a high correlation appears on the screen 21 disposed on the rear focal plane of the second Fourier transform lens 20. Based on the position of this spot, the position of the same pattern as the pattern to be extracted in the input image can be detected and recognized and classified. A hologram is used as the matched filter 41. However, if a hologram is used as described above, it is necessary to create a separate filter for each pattern to be extracted, and there is a problem in filter alignment.

In the apparatus of the present invention, the extraction patterns are further classified into smaller categories in the subsequent processing. Here, FIG. 3 shows an example of an input image, FIG. 4 shows an example of a representative pattern of the input image of FIG. 3, FIG. 5 shows an example of an output image of the input image of FIG.
FIG. 6 shows an example of an extraction pattern extracted by analyzing the output image.

On the other hand, in FIG. 1, the second light flux 12b of the two light fluxes split by the half mirror 14 is deflected by the mirror 24 and enters the second image display means 25. The reference pattern stored in the computer 18 is drawn in an intensity transmittance distribution on the second image display means 25 disposed on the front focal plane of the third Fourier transform lens 26, and the second light flux 12b is transmitted to form the reference pattern. Is read. Second
The light flux 12b passes through the third Fourier transform lens 26,
A Fourier transform image of the reference pattern is formed on the second spatial light modulator 27 disposed on the rear focal plane of the third Fourier transform lens. In the second spatial light modulator 27, the intensity pattern of one Fourier transform image among the extracted patterns output by the above-described processing output by the second Fourier transform image output means 28 is displayed as a transmittance distribution. Therefore, the Fourier transform image of the reference pattern is modulated by the Fourier transform image of the extraction pattern when passing through the second spatial light modulator 27, and the calculation of the product of the corresponding spatial frequency components of the reference pattern and the extraction pattern is performed. Done.

Thereafter, the second light flux 12b passes through the fourth Fourier transform lens 29 and is subjected to Fourier transform again, so that the second light flux 12b falls on the light receiving surface of the light receiving means 30 disposed on the front focal plane of the fourth Fourier transform lens 29. A pattern is created as a result of the cross-correlation operation between the reference pattern and the extracted pattern. To be precise, this pattern is obtained by performing a combination operation of an extracted pattern in addition to a correlation operation between the extracted pattern and the reference pattern.
The light quantity of this pattern is detected by the light receiving means 30, and the output of the cross-correlation calculation is obtained. Hereinafter, the second image display means 25
Are sequentially changed, and the above processing is performed to obtain an output of a cross-correlation operation for a plurality of reference patterns.

These outputs are input to the identification / judgment device 31 to judge which small category the extracted pattern belongs to. This determination is realized by checking in advance what range of values the plurality of outputs of the extracted pattern for each reference pattern indicate. As long as the ranges do not overlap in the output for all reference patterns among a plurality of sub-categories, the binary logic of whether or not the output value for each reference pattern of the input extracted pattern falls within the range checked beforehand. It can be done from a combination. In addition, Mahalanobis distance, city area distance, and the like have been proposed as a method for statistically evaluating a distance from each small category of a vector (referred to as a feature vector) having a feature parameter of an input extracted pattern as a component. The determination can be made using these distances. The identification determination device 31 can realize the determination function by software using the computer 18 or can realize the same function by hardware using an electronic circuit.

As the first and second image display means 15 and 25 used in the apparatus of the present invention, a liquid crystal panel or the like, which is a general electric address type spatial light modulator, can be used. Here, the first and second spatial light modulators 17,
A liquid crystal panel or the like, which is a general electric address type spatial light modulator, can be used for 27. Further, a CCD camera, an imaging tube, or the like can be used as the imaging device 22.

Further, a process of arranging a circular mask or the like having an appropriate radius centered on the optical axis on the first spatial light modulator 17 to cut off the representative pattern, the input image, or both images near the optical axis. May be performed. As a result, low spatial frequency components which are almost common to any pattern are cut, and a pattern having a correlation with the representative pattern can be more efficiently extracted. Other than using a mask, there is a method of removing an image near the optical axis when outputting a Fourier transform image of the correlation pattern by the Fourier transform image output device 19.

Further, when a binary image of the intensity pattern of the Fourier transform image of the representative pattern is displayed on the first spatial light modulator 17, the Fourier transform image of the input image that matches the Fourier transform image of the representative pattern can be obtained. Since the portion is transmitted almost without being affected by the spatial light modulator, it is possible to prevent a decrease in the contrast of a pattern having a correlation with the representative pattern remaining in the output image. Also, the second spatial light modulator 2
If a binarized image of the Fourier transform image of the extracted pattern is displayed in FIG. 7, the portion of the Fourier transform image of the reference pattern that matches the Fourier transform image of the extracted pattern is transmitted without being affected by the spatial light modulator. Light receiving means 3
It is possible to prevent a decrease in the light amount of the light flux reaching zero.

As such a spatial light modulator, an electric address type using a ferroelectric liquid crystal can be used. In particular, a spatial light modulator using a ferroelectric liquid crystal is suitable not only for binarization of an image but also for its high resolution and high contrast in principle. In addition, if the binarization process is performed using an image processing device or the like, the same can be achieved using a general electric address type spatial light modulator.

As the light receiving means 30, an image pickup device such as a photodiode or a CCD can be used. Further, in the present embodiment, the light beam is divided into two beams by using the half mirror 14 and the mirror 24. There is no problem.

Embodiment 2 FIG. 7 is a schematic diagram showing another embodiment of the optical pattern recognition / classification apparatus of the present invention. In the second embodiment, optically obtaining the Fourier transform image of the representative pattern displayed on the first spatial light modulator 17 and the Fourier transform image of the extracted pattern displayed on the second spatial light modulator 27, The pattern is displayed in parallel on the second image display means 25, and the first lens is disposed at a position corresponding to the reference pattern.
And the use of the second Fourier transform lens arrays 51 and 52. Also in the present embodiment, a pattern belonging to a large category including several small categories is extracted from an input image by the pattern extracting means 1, and an extraction belonging to the extracted large category is performed by the optical correlation calculating means 2 and the discriminating means 3. A detailed classification of which small category the pattern belongs to is performed.

The coherent light beam 12 emitted from the laser 11 is expanded to an appropriate light beam diameter by a beam expander 13 and collimated.
b, 14c, and 14d, respectively, and the luminous flux 12a, 1
2b, 12c, and 12d. The light flux 12c enters the image display means 53. On the image display means 53 arranged on the front focal plane of the Fourier transform lens 54, the intensity pattern of the representative pattern stored in the computer 18 is drawn as a transmittance distribution, and the light beam 12c is transmitted and the representative pattern is displayed. Is read. Thereafter, the light flux 12c passes through the Fourier transform lens 54, and forms a Fourier transform image of the representative pattern on a screen 55 arranged on the rear focal plane. This image is captured by the imaging unit 56. Here, as the representative pattern displayed on the image display means 53, the representative pattern may be directly imaged by another imaging means and the image may be displayed.

On the other hand, the light beam 12a is
5 is incident. On the first image display means 15 arranged on the front focal plane of the first Fourier transform lens 16, an input image from which a pattern belonging to a large category is to be extracted is drawn in a transmittance distribution of an intensity pattern, and the light flux 12a is The input image is read through. The light beam 12a passes through the first Fourier transform lens 16 and is
A Fourier transform image of the input image is formed on the first spatial light modulator 17 disposed on the rear focal plane of the input image. First spatial light modulator 17
Since the Fourier transform image of the representative pattern imaged by the imaging means 56 is displayed as the transmittance distribution of the intensity pattern, when the light flux 12a is transmitted, the modulation is performed by the intensity pattern of the Fourier transform image of the representative pattern. Then, the product of the corresponding spatial frequency components of the input image and the representative pattern is calculated. After this, the light flux 12a is transmitted through the second Fourier transform lens 20 and subjected to Fourier transform again, and on a screen 21 arranged on the front focal plane of the second Fourier transform lens 20, a cross-correlation calculation of the input image and the representative pattern is performed. The result is output.

More specifically, the output image is an image in which the input image having a low correlation with the representative pattern is deleted.
The part of the input image that has a high degree of correlation with the representative pattern is
The part that is output without lowering the contrast and, conversely, has little correlation with the representative pattern is output with little contrast, and an image in which a pattern similar to the representative pattern remains is output. The image is taken by the imaging means 22. After the output image is captured by the image capturing unit 22, the image is displayed again on the image display unit 15 as an input image, and the above-described series of processing is repeated an appropriate number of times. Disappear,
An image in which only a pattern having a higher similarity to the representative pattern remains until the end is output. At this time, the number of repetitions is controlled to an appropriate number by a program of the computer 18.

Finally, the output image is taken into the computer 18 through the image pickup means 22, and the position of the pattern remaining in the output image is detected by the image processing device 23. A pixel having an output value equal to or more than a certain value in the output image is detected, a group of the pixels having a certain number or more is detected, and information on the position is output. Based on this position information, a region of an appropriate size corresponding to the size of the group of pixels is cut out from the input image to form an extraction pattern. The image processing device 23 can be realized using a general image processing board or the like used by being connected to a personal computer or the like.

By the above-described processing, one or a plurality of patterns correlated with the target pattern are extracted and become extracted patterns. That is, the patterns belonging to the large category having the common feature are all extracted from the input image as the extraction patterns. The extraction patterns are further classified into smaller categories in the subsequent processing.

In FIG. 7, on the other hand, the light flux 12d enters the image display means 58. On the image display means 58 disposed on the front focal plane of the Fourier transform lens 59, one of the extracted patterns extracted by the above processing is drawn as a transmittance distribution, and the light flux 12d is transmitted therethrough. The extraction pattern is read. Thereafter, the light beam 12d passes through the Fourier transform lens 59, and forms a Fourier transform image of the extracted pattern on a screen 60 arranged on the rear focal plane. This image is captured by the imaging device 61. The extracted Fourier-transformed image is input to the image processing device 62, and is converted by the Fourier-transform lens array 51 into a plurality of images arranged in parallel at positions corresponding to the respective lenses. The image processing device 62 can be realized using a general image processing board or the like used by connecting to a personal computer or the like.

Further, the light beam 12b is deflected by the mirror 24 and enters the second image display means 25. On the second image display means 25 arranged on the front focal plane of each lens of the Fourier transform lens array 51, a plurality of reference patterns stored in the computer 18 are drawn in parallel with an intensity transmittance distribution,
The plurality of reference patterns are read through the light flux 12b. The light flux 12b passes through the Fourier transform lens array 51 in which each lens is arranged at a position corresponding to each of the reference patterns, and falls on the second spatial light modulator 27 arranged on the rear focal plane of each lens. The Fourier transform images of the plurality of reference patterns can be formed at positions corresponding to the respective lenses. Since the parallel image of the Fourier transform image of the extraction pattern output from the image processing device 62 is displayed on the second spatial light modulator 27, the Fourier transform image of each of the plurality of reference patterns is converted to the second spatial light modulation. When the light passes through the detector 27, it is modulated by the Fourier transform image of the extracted pattern, and the product of the corresponding spatial frequency components of the reference pattern and the extracted pattern is calculated.

Thereafter, the light beam 12b is again Fourier-transformed by each lens of the Fourier transform lens array 52 in which each lens is arranged corresponding to the position of each of the plurality of reference patterns. The light enters the light receiving means array 63 in which the light receiving surface is disposed at the front focal position, and a pattern resulting from the cross-correlation calculation of each reference pattern and the extracted pattern is formed on each light receiving surface. The light amounts of these patterns are subjected to photoelectric conversion and the like by respective light receiving means of the light receiving device array 63, and a plurality of outputs of a cross-correlation operation between each reference pattern and the extracted pattern are simultaneously detected.

These outputs are input to the identification / judgment device 31 to judge which small category the extracted pattern belongs to. This determination is realized by checking in advance what range of values the plurality of outputs of the extracted pattern for each reference pattern indicate. As long as the ranges do not overlap in the output for all reference patterns among a plurality of sub-categories, the binary logic of whether or not the output value for each reference pattern of the input extracted pattern falls within the range checked beforehand. It can be done from a combination. In addition, as a method of statistically evaluating the distance from each subcategory of a vector (referred to as a feature vector) having a feature parameter of an input extracted pattern as a component, determination is performed using a Mahalanobis distance or a city area distance. You can also. The identification determination device 31 includes the computer 1
The determination function can be realized by software using the software 8, or the same function can be realized by hardware using an electronic circuit.

The image display means 15, 25, 5 used here
A liquid crystal panel or the like, which is a general electric address type spatial light modulator, can be used for 3, 58. Here, as the spatial light modulators 17 and 27, a liquid crystal panel or the like, which is a general electric address type spatial light modulator, can be used. In the spatial light modulators 17 and 27, a representative pattern or a Fourier transform image of an extracted pattern is written as a gray level or a binary image, but when written as a binary image, a representative pattern or a light receiving device in the output image is written. Array 62
The effect of suppressing a decrease in the contrast of the reference pattern incident on the pixel is obtained. In such a case, it is preferable that the image is binarized using an image processing device or the like.

Further, the imaging devices 22, 56 and 61 include C
A CD camera or an image pickup tube can be used. In addition,
In this embodiment, the light flux 12 is divided into half mirrors 14a, 14a.
Although it is divided into four according to b and 14c, there is no problem even if an apparatus that prepares four laser beams and four beam expanders and separately generates four light beams is used.

Embodiment 3 FIG. 8 is a schematic diagram showing still another embodiment of the optical pattern recognition / classification apparatus of the present invention. The device shown in FIG. 8 is characterized in that an optical writing type device typified by a liquid crystal light valve is used as the spatial light modulator. Also in the third embodiment, a pattern belonging to a large category including some small categories is extracted from an input image by a pattern extracting unit 1, and an optical correlation calculating unit 2 and a discriminating unit 3 are used.
Thus, detailed classification of which small category the extracted pattern belonging to the extracted large category belongs to is performed.

The coherent light beam 72 emitted from the semiconductor laser 71 is collimated by the collimator lens 73 and enters the image display means 53. On the image display means 53 arranged on the front focal plane of the Fourier transform lens 54, the intensity pattern of the representative pattern stored in the computer 18 is drawn as a transmittance distribution, and the light beam 72 is transmitted and the representative pattern is displayed. Is read. Thereafter, the light beam 72 passes through the Fourier transform lens 54, and forms a Fourier transform image of the representative pattern on the writing surface of the light writing type spatial light modulator 74 arranged on the rear focal plane. Accordingly, in the optical writing type spatial light modulator 74, for example, when a liquid crystal light valve is used, the resistance of the photoconductor layer in the light valve changes according to the light intensity distribution of the incident Fourier transform image, so that the liquid crystal An electric field is applied to the liquid crystal molecules to change the arrangement of liquid crystal molecules, and a Fourier transform image of a representative pattern is written.

On the other hand, the coherent light flux 76 emitted from the semiconductor laser 75 is collimated by the collimator lens 77 and enters the image display means 15. On the image display means 15 disposed on the front focal plane of the Fourier transform lens 16, the input image is drawn in the transmittance distribution of the intensity pattern, and the light flux 76 is transmitted to read the input image. The modulated light flux 76 is reflected by the polarization beam splitter 78 and enters the Fourier transform lens 16. At this time, in order to maximize the reflectance at the polarization beam splitter 78, the light emitted from the image display means 15 should be in a linear polarization state (usually an s-polarization state) corresponding to the total reflection of the polarization beam splitter. After passing through the Fourier transform lens 16, the Fourier transform image of the input image is incident on the readout surface of the optical writing type spatial light modulator 74 arranged on the rear focal plane and is reflected. Optical modulator 7
4, the Fourier transform image of the representative pattern is written as described above, and the structure of the liquid crystal is changed. Therefore, the portion of the Fourier transform image of the input image that hits the portion of the Fourier transform image of the representative pattern is light Write-type spatial light modulator 74
The state of polarization changes in accordance with the pattern written in, and the reflected light again passes through the Fourier transform lens 16 and again enters the polarization beam splitter 78. Here, a linearly polarized light component changed in accordance with the Fourier transform image of the target pattern written in the optical writing type spatial light modulator 74 and a component (usually a p-polarized light component) orthogonal to that before the change is transmitted to the polarization beam splitter 78. Transmitted as it is, Fourier transform lens 16
Reaches the screen 21 disposed on the front focal plane of the camera, and this image becomes an output image. Here, the Fourier transform lens 16
Is the Fourier transform lens 20 in the first or second embodiment.
And the function of two lenses is realized by one lens.

This output image is the result of the cross-correlation operation between the input image and the representative pattern. Specifically, the output image is an image in which the input image having a low correlation with the representative pattern is removed, and a portion of the input image having a high correlation with the representative pattern is output without lowering the contrast. Conversely, a portion having little correlation with the representative pattern is output with little contrast, and an image in which a pattern similar to the representative pattern remains is output. . After the output image is captured by the image capturing unit 22, the image is displayed again on the image display unit 15 as an input image, and the above-described series of processing is repeated an appropriate number of times. An image that disappears and only the pattern having a higher similarity to the representative pattern remains until the end is output. At this time, the number of repetitions is controlled to an appropriate number by a program of the computer 18.

Finally, the output image is taken into the computer 18 via the imaging means 22, and the position of the pattern remaining in the output image is detected by the image processing device 23. A pixel having an output value equal to or more than a certain value in the output image is detected, a group of the pixels having a certain number or more is detected, and information on the position is output. Based on this position information, a region of an appropriate size corresponding to the size of the group of pixels is cut out from the input image to form an extraction pattern. The image processing device 23 can be realized using a general image processing board or the like used by being connected to a personal computer or the like. By the above-described processing, one or a plurality of patterns correlated with the target pattern are extracted and become extracted patterns. That is, the patterns belonging to the large category having the common feature are all extracted from the input image as the extraction patterns. The extraction patterns are further classified into smaller categories in the subsequent processing.

Further, the coherent light beam 80 emitted from the semiconductor laser 79 is collimated by the collimator lens 81 and enters the image display means 58. On the image display means 58 arranged on the front focal plane of the Fourier transform lens 59, one intensity pattern of the extracted patterns extracted by the above processing is drawn as a transmittance distribution, and the light beam 80 is transmitted therethrough. The extraction pattern is read.
Thereafter, the light beam 80 is transmitted through the Fourier transform lens 59, and the light writing type spatial light modulator 8 disposed on the rear focal plane is provided.
The Fourier transform image of the extracted pattern is formed on the writing surface of No. 2 and the Fourier transform image of the extracted pattern is written in the optical writing type spatial light modulator 82 by the above operation.

On the other hand, the coherent light beam 84 emitted from the semiconductor laser 83 is collimated by the collimator lens 85 and enters the image display device 25. One or more reference patterns are drawn in the transmittance distribution of the intensity pattern on the image display means 25 arranged on the front focal plane of the Fourier transform lens 26, and the reference pattern is read out through the light beam 84. The modulated light beam 84 is reflected by the polarization beam splitter 86 and enters the Fourier transform lens 26. At this time, in order to maximize the reflectance at the polarization beam splitter 86, the light emitted from the image display device 25 may be set to a linear polarization state (normally, an s-polarization state) corresponding to the total reflection of the polarization beam splitter. After passing through the Fourier transform lens 26, the Fourier transform image of one or a plurality of reference patterns is incident on the readout surface of the optical writing type spatial light modulator 82 arranged on the rear focal plane and is reflected. Since the Fourier transform image of the extracted pattern is written in the optical writing type spatial light modulator 82 as described above and the structure of the liquid crystal is changed, the reference pattern corresponding to the part of the Fourier transformed image of the extracted pattern is changed. The polarization state of the portion of the Fourier transform image changes in accordance with the pattern written to the optical writing type spatial light modulator 82, and the reflected light again passes through the Fourier transform lens 26 and again enters the polarization beam splitter 86. .

Here, the linearly polarized light changed in accordance with the Fourier transform image of the extracted pattern written in the optical writing type spatial light modulator 82, and a component orthogonal to the change (usually a p-polarized light component) is a polarized light beam. The light-receiving unit array 63 having the light-receiving surface disposed on the front focal plane of the Fourier transform lens 26 at a position corresponding to one or a plurality of reference patterns drawn on the image display unit 25 and transmitted through the splitter 86 as it is. Then, a pattern is formed on the light receiving surface as a result of the cross-correlation operation of the extracted pattern and one or more reference patterns. The light receiving means array 6
The photoelectric conversion is performed by each of the light receiving means 3 and a plurality of outputs of a cross-correlation operation between one or a plurality of reference patterns and the extracted patterns are simultaneously detected. Here, the Fourier transform lens 26 is used in the first embodiment (FIG. 1) or the second embodiment.
The function of the Fourier transform lens 29 in FIG. 7 is also used, and the function of two lenses is realized by one lens.

These outputs are input to the identification / judgment device 31 to judge which small category the extracted pattern belongs to. This determination is realized by checking in advance what range of values the plurality of outputs of the extracted pattern for each reference pattern indicate. As long as the ranges do not overlap in the output for all reference patterns among a plurality of sub-categories, the binary logic of whether or not the output value for each reference pattern of the input extracted pattern falls within the range checked beforehand. It can be done from a combination. In addition, as a method of statistically evaluating the distance from each subcategory of a vector (referred to as a feature vector) having a feature parameter of an input extracted pattern as a component, determination is performed using a Mahalanobis distance or a city area distance. You can also. The identification determination device 31 includes the computer 1
The determination function can be realized by software using the software 8, or the same function can be realized by hardware using an electronic circuit.

The image display means 15, 2 used in the apparatus of FIG.
For 5, 53, 58, a liquid crystal panel or the like, which is a general electric address type spatial light modulator, can be used. Further, a CCD camera, an image pickup tube or the like can be used as the image pickup means 22. In the device of the third embodiment (FIG. 8), an optical writing type spatial light modulator is used. As such a spatial light modulator, a liquid crystal panel combined with a photoconductive layer is generally used. , And those using nematic-type liquid crystal are on sale.

A Fourier transform image of a representative pattern or an extracted pattern is written in the optical writing type spatial light modulators 74 and 82 as a gray level or a binary image. The effect of suppressing a decrease in the contrast of the extracted pattern or the reference pattern incident on the light receiving device array 63 is obtained. As such a spatial light modulator, a device using a ferroelectric liquid crystal can be used. In particular, a spatial light modulator using a ferroelectric liquid crystal is preferable because it not only binarizes an image but also has a high resolution and a high contrast in principle.

Further, as an optical address type spatial light modulator, BSO (bismuth silicon oxide; Ba1
A medium using a medium having a photoinduced refractive index change such as (2SiO20) can also be used. In particular, in the case of BSO material, using visible light from a short wavelength having high sensitivity as writing light,
Writing and reading can be performed simultaneously by using light in a wavelength range from red having low writing sensitivity to an infrared region as reading light. However, the BSO material can be used in a reflection type by using a reflection plate, but is usually used as a transmission type in many cases. In this case, it is necessary to change the incident direction of the reading light incident on the BSO material, but this is not an essential means of the present invention.

Fourth Embodiment In the fourth embodiment, the processing by the feature amount detecting means and the discriminating means is the same as in the first embodiment (FIG. 1), the second embodiment (FIG. 7) and the third embodiment (FIG. 8). Therefore, only the pattern extracting device will be described. FIG. 9 is a schematic configuration diagram of a portion of a pattern extraction device of still another embodiment of the optical pattern recognition / classification device of the present invention. The fourth embodiment is characterized in that the image feedback means is realized only by optical processing.

In FIG. 9, a coherent light beam 72 emitted from a semiconductor laser 71 is collimated by a collimator lens 73 and enters an image display means 53. On the image display means 53 arranged on the front focal plane of the Fourier transform lens 54, the intensity pattern of the representative pattern stored in the computer 18 is drawn as a transmittance distribution, and the light beam 72 is transmitted and the representative pattern is displayed. Is read. Thereafter, the light beam 72 passes through the Fourier transform lens 54, and forms a Fourier transform image of the representative pattern on the writing surface of the light writing type spatial light modulator 74 arranged on the rear focal plane. Thereby, the optical writing type spatial light modulator 74
For example, when a liquid crystal light valve is used, an electric field is applied to the liquid crystal due to a change in the resistance of the photoconductor layer in the light valve according to the incident light intensity distribution of the Fourier transform image, and the arrangement of the liquid crystal molecules changes. Then, a Fourier transform image of the representative pattern is written.

On the other hand, in the initial state, the mirror 91 is arranged as shown in FIG. 10, and the light beam from the input image 90 passes through the coupling lens 92 and is placed on the light writing type spatial light modulator 93. An image of the input image 90 is formed. Thus, in the light-writing type spatial light modulator 93, for example, when a liquid crystal light valve is used, the resistance in the photoconductor layer in the light valve changes according to the light intensity distribution of the incident Fourier transform image. An electric field is applied to the liquid crystal molecules to change the arrangement of liquid crystal molecules, and an image of an input image is written. The input image may be a natural image such as an object or a daily scene.

Further, the coherent light beam 76 emitted from the semiconductor laser 75 is collimated by the collimator lens 77, passes through the polarization beam splitter 94, and enters the optical writing type spatial light modulator 93. At this time, in order to make the transmittance of the polarizing beam splitter 94 the best, the light emitted from the semiconductor laser 75 is linearly polarized (usually p-polarized) corresponding to the total transmission of the polarizing beam splitter.
What should I do?

After passing through the polarizing beam splitter 94,
The light flux 76 is applied to the read surface of the
At this time, the input image is written into the optical writing spatial light modulator 93 as described above, and the structure of the liquid crystal is changed. The polarization state of the portion changes in accordance with the pattern written in the optical writing spatial light modulator 93, and the reflected light enters the polarization beam splitter 94 again. Here, a linearly polarized light component changed in accordance with the image of the input image written in the optical writing type spatial light modulator 93 and a component (usually s is a polarized light component) orthogonal to that before the change is reflected by the polarization beam splitter 94. , Through the half mirror 95 and the mirror 96,
The light enters the polarization beam splitter 78, is reflected similarly to the case of the polarization beam splitter 94, and enters the Fourier transform lens 16. Here, the half mirror 95 is used to capture an output image.

After passing through the Fourier transform lens 16, the Fourier transform image of the input image is incident on the readout surface of the optical writing type spatial light modulator 74 arranged on the rear focal plane and is reflected. As described above, the Fourier transform image of the representative pattern is written in the spatial light modulator 74, and the structure of the liquid crystal is changed. Therefore, the Fourier transform image of the input image corresponding to the Fourier transform image of the representative pattern is obtained. The polarization state of the portion changes in accordance with the pattern written in the optical writing type spatial light modulator 74, and the reflected light passes through the Fourier transform lens 16 again, and again enters the polarization beam splitter 78.

Here, the linearly polarized light changed in accordance with the Fourier transform image of the target pattern written in the optical writing type spatial light modulator 74, and the component orthogonal to the change (usually the p-polarized light component) is a polarized beam. The light passes through the splitter 78 as it is. Here, the Fourier transform lens 16 corresponds to the first embodiment.
(FIG. 1) or the function of the Fourier transform lens 20 in the second embodiment (FIG. 7) is also used, and the function of two lenses is realized by one lens.

The transmitted image is an output image obtained by performing a cross-correlation operation between the input image and the representative pattern. Specifically, the output image is an image in which the input image having a low correlation with the representative pattern is removed, and a portion of the input image having a high correlation with the representative pattern is output without lowering the contrast. Conversely, a portion having little correlation with the representative pattern is output with little contrast, and an image in which a pattern similar to the representative pattern remains is output. With the mirror 91 arranged as shown in FIG. 9, the output image is made incident on the light-writing type spatial light modulator 93 arranged at the front focal position of the Fourier transform lens 16, and the output image is written again as an input image. If the above series of processing is repeated an appropriate number of times, a portion having a low correlation with the representative pattern in the input image disappears, and an image in which only a pattern having a similarity with the representative pattern remains to the end is output.

The final output image is written to the optical writing type spatial light modulator 93, and the image is read out in the same manner as described above. I do. With this imaging lens 97,
An output image is formed on the screen 21. This image is picked up by the image pickup means 22 and taken into the computer 18. After that, the position of the pattern remaining in the output image is detected by the image processing device 23, and an area is cut out from the input image based on the position information to become an extracted pattern. That is, the patterns belonging to the large category having the common feature are all extracted from the input image as the extraction patterns. The extraction patterns are further classified into smaller categories in the subsequent processing. Subsequent feature amount detecting means and discriminating means are the same as those in the first to third embodiments, and will not be described.

As the image display means 53 used in the apparatus shown in FIG. 9, a liquid crystal panel or the like, which is a general electric address type spatial light modulator, can be used. Further, the imaging unit 22
, A CCD camera, an image pickup tube or the like can be used. In the fourth embodiment, an optical writing type spatial light modulator is used. However, as such a spatial light modulator, a liquid crystal panel combined with a photoconductive layer is generally used, and a nematic-type liquid crystal is used. What has been released.

The Fourier transform image and the input image of the representative pattern are written in the optical writing type spatial light modulators 74 and 93 in the form of gray level or binary image. The effect of suppressing a decrease in contrast of the extracted pattern or the like is obtained. As such a spatial light modulator, a device using a ferroelectric liquid crystal can be used. In particular, a spatial light modulator using a ferroelectric liquid crystal is preferable because it not only binarizes an image but also has a high resolution and a high contrast in principle.

Further, as the optical address type spatial light modulator, BSO (bismuth silicon oxide; Ba1
A medium using a medium having a photoinduced refractive index change such as (2SiO20) can also be used. In particular, in the case of BSO material, using visible light from a short wavelength having high sensitivity as writing light,
Writing and reading can be performed simultaneously by using light in a wavelength range from red having low writing sensitivity to an infrared region as reading light. However, the BSO material can be used in a reflection type by using a reflection plate, but is usually used as a transmission type in many cases. In this case, it is necessary to change the incident direction of the reading light incident on the BSO material, but this is not an essential means of the present invention.

Fifth Embodiment In the fifth embodiment, the processes of the feature amount detecting means and the discriminating means are the same as those in the first to third embodiments, and therefore only the pattern extracting device will be described. FIG. 11 is a schematic configuration diagram of a part of a pattern extraction device of still another embodiment of the optical pattern recognition / classification device of the present invention. In the fifth embodiment,
Another characteristic is that the image feedback means is realized only by optical processing.

In FIG. 11, a coherent light beam 72 emitted from a semiconductor laser 71 is
And is incident on the image display means 53.
The intensity pattern of the representative pattern stored in the computer 18 is drawn as a transmittance distribution on the image display means 53, and the light beam 72 is transmitted and the representative pattern is read. Thereafter, the light beam 72 passes through the Fourier transform lens 54, and forms a Fourier transform image of the representative pattern on the writing surface of the optical writing type spatial light modulator 74 arranged on the focal plane. Thereby, the optical writing type spatial light modulator 74
For example, when a liquid crystal light valve is used, an electric field is applied to the liquid crystal due to a change in the resistance of the photoconductor layer in the light valve according to the incident light intensity distribution of the Fourier transform image, and the arrangement of the liquid crystal molecules changes. Then, a Fourier transform image of the representative pattern is written.

On the other hand, the coherent light beam 76 emitted from the semiconductor laser 75 is collimated by the collimator lens 77 and enters the image display means 15. The input image is drawn on the image display means 15 with the transmittance distribution of the intensity pattern, and the light flux 76 is transmitted to read the input image.
The modulated light flux 76 passes through the polarization beam splitter 94, is reflected by the mirror 96, and is written on the front focal plane of the Fourier transform lens 16 by the light writing type spatial light modulator 93.
Incident on. At this time, in order to make the transmittance of the polarization beam splitter 94 the best, the light emitted from the image display device 15 should be in a linear polarization state (usually a p-polarization state) corresponding to the total transmission of the polarization beam splitter.

In the initial state, the semiconductor laser 10
1 emits light and is collimated by a collimator lens 102 to form a half mirror 95 and a Dove prism 103.
After that, light is incident on the entire writing surface of the optical writing type spatial light modulator 93. Thus, for example, in the case of a liquid crystal light valve, an electric field is applied to all the liquid crystals of the photoconductor layer in the light valve, the arrangement of the liquid crystal molecules changes, and the entire surface is written. At this time, instead of the semiconductor laser 101, an incoherent light source such as a metal halide lamp may be used. In addition to the whole surface writing by light, the same effect can be obtained by changing the driving state of the optical writing type spatial light modulator 93 so that the state of the liquid crystal changes even without writing light.

The light beam 76 incident on the reading surface of the optical writing type spatial light modulator 93 is reflected and enters the polarizing beam splitter 94 again via the mirror 96. At this time, the structure of the liquid crystal over the entire surface is changed. Because of the change, the polarization state of the light beam incident on the readout surface of the modulator changes to a polarization (usually, s-polarization) orthogonal to that before the change, and is reflected by the polarization beam splitter 94 to produce another polarization. The light enters the beam splitter 78. That is, the input image displayed on the image display device reaches the polarizing beam splitter 78 as it is.
Also in the polarization beam splitter 78, the light flux 76 is reflected in the same manner as described above, enters the Fourier transform lens 16, and transmits. After passing through the Fourier transform lens 16, the Fourier transform image of the input image is incident on the readout surface of the optical writing type spatial light modulator 74 arranged on the rear focal plane, and is reflected.

At this time, the Fourier transform image of the representative pattern is written in the optical writing type spatial light modulator 74 as described above, and the structure of the liquid crystal is changed. In the part of the Fourier transform image of the input image that has hit, the polarization state changes according to the pattern written to the optical writing type spatial light modulator 74,
The reflected light again passes through the Fourier transform lens 16 and again enters the polarization beam splitter 78. Here, a linearly polarized light component changed in accordance with the Fourier transform image of the representative pattern written in the optical writing type spatial light modulator 74 and a component (usually a p-polarized light component) orthogonal to the pre-change linear polarization component passes through the polarization beam splitter 78. The light passes through the half mirror 95 and the Dove prism 103, passes through the half mirror 95, and passes through the Dove prism 103.
The output image is incident on the writing surface of. Here, the half mirror 95 is used to capture an output image, and the Dove prism is used to correct the vertical or horizontal reversal of the image.

At this time, if the emission of the semiconductor laser 101 is stopped, the light beam incident on the optical writing type spatial light modulator 93 becomes only the above-mentioned output image, and the output image is written according to the principle described above. Here, in the case where the entire writing is initially performed only by controlling the driving method of the modulator without using light, the driving state changes the state of the liquid crystal in the portion where the writing light is incident. By changing to such normal driving, an output image is similarly written.

This output image is an image obtained by performing a cross-correlation operation between the input image and the representative pattern. Specifically, an image having a low correlation with the representative pattern in the input image is a deleted image, and a portion having a high degree of correlation with the representative pattern in the input image is output without lowering the contrast, and The portion having little correlation with the representative pattern is output with little contrast, and an image in which a pattern similar to the representative pattern remains is output.

Thereafter, the light flux 76 modulated by the image display device 15 with the intensity pattern of the input image is incident on and reflected by the light-writing type spatial light modulator 93. At this time, the light-writing type spatial light modulator 93 Since the output image is written as described above and the structure of the liquid crystal is changed, the portion of the input image that hits the portion of the written output image is written to the optical writing type spatial light modulator 93. The polarization state changes in accordance with the pattern, and then enters the polarization beam splitter 94 again via the mirror 96. Here, a component (usually, an s-polarized component) of the linearly polarized light changed according to the output image written in the optical writing type spatial light modulator 93 and orthogonal to that before the change is reflected by the polarization beam splitter 94. Thereafter, the light is reflected by the polarizing beam splitter 78 as described above,
The light enters the Fourier transform lens 16.

The pattern incident on the Fourier transform lens 16 is an image in which, as in the case of the output image, a pattern having a small correlation with the representative pattern in the input image is removed, and the degree of correlation with the representative pattern in the input image is reduced. Are output without a decrease in contrast, and conversely, portions with little correlation with the representative pattern are output with little contrast, and are images in which a pattern similar to the representative pattern remains. .

Therefore, similarly, if the above-described series of processing is repeated an appropriate number of times, portions having a low correlation with the representative pattern in the input image disappear, and only a pattern having a higher similarity with the representative pattern remains at the end. The remaining image is output.

The final output image is partially deflected by the half mirror 95 to form an output image on the screen 21 disposed on the front focal plane of the Fourier transform lens 16. This image is picked up by the image pickup device 22 and taken into the computer 18. Thereafter, the image processing device 23
, The position of the pattern remaining in the output image is detected, an area is cut out from the input image based on the position information, and an extracted pattern is obtained. That is, the patterns belonging to the large category having the common feature are all extracted from the input image as the extraction patterns. The extraction patterns are further classified into smaller categories in the subsequent processing. Subsequent feature amount detecting means and discriminating means are the same as those in the first to third embodiments, and will not be described.

The image display means 15, used in the apparatus of FIG.
A liquid crystal panel or the like, which is a general electric address type spatial light modulator, can be used for 53. Further, as the imaging device 22, a CCD camera, an imaging tube, or the like can be used. In the fifth embodiment, the optical writing type spatial light modulator is used. However, as such a spatial light modulator, a liquid crystal panel combined with a photoconductive layer is generally used.
A product using a nematic-type liquid crystal is on the market.

The Fourier transform image or the output image of the representative pattern is written in the optical writing type spatial light modulators 74 and 93 in the form of a gray level or a binary image. The effect of suppressing a decrease in contrast of the extracted pattern or the like is obtained. As such a spatial light modulator, a device using a ferroelectric liquid crystal can be used. In particular, a spatial light modulator using a ferroelectric liquid crystal is preferable because it not only binarizes an image but also has a high resolution and a high contrast in principle.

Further, as the optical address type spatial light modulator, BSO (bismuth silicon oxide; Ba1
A medium using a medium having a photoinduced refractive index change such as (2SiO20) can also be used. In particular, in the case of BSO material, using visible light from a short wavelength having high sensitivity as writing light,
Writing and reading can be performed simultaneously by using light in a wavelength range from red having low writing sensitivity to an infrared region as reading light. However, the BSO material can be used in a reflection type by using a reflection plate, but is usually used as a transmission type in many cases. In this case, it is necessary to change the incident direction of the reading light incident on the BSO material, but this is not an essential means of the present invention.

Embodiment 6 This embodiment 6 is the same as the above-described embodiments 1 to 5 up to the point where the feature parameter is obtained by using the feature amount detecting means such as an optical correlation calculation device. Only the part of the device 31 will be described. In the sixth embodiment, a fuzzy logic operation device is used for the identification determination device 31. This is because, after creating a membership function based on a plurality of feature amounts of the extracted pattern belonging to the category obtained by the feature amount detecting means, the feature amount of the unknown extracted pattern is compared with the membership function described above. A membership value is obtained for each feature parameter, and an extraction pattern is determined based on a plurality of obtained membership values.

First, the characteristic amount is acquired by the characteristic amount detecting means. In order to reduce the influence when the characteristic amount detecting means has an unstable output, it is preferable to perform the extraction plural times for each extracted pattern. Therefore, the average value and the standard deviation amount, which are one of the representative values of the output values of the characteristic parameters, are calculated from the data of a plurality of times, and the membership function for the output intensity is set to a trapezoidal shape. The range up to the range is defined as a membership value of 1, a magnitude three times the standard deviation is defined as a membership value of 0, and a portion corresponding to the hypotenuse of the trapezoid is defined as being equal to three times the standard deviation.

FIGS. 12 to 14 are membership function graphs of a small category to which pattern A to which the output of feature parameter 1 and feature parameter 2 of a certain extracted pattern A is expressed as a membership value. In this way, membership functions for each division pattern are created for all the small categories and stored in the identification and classification device 31. Next, an unknown extracted pattern is shown in place of the pattern A, the output of each component of the feature parameter is obtained, and the output is compared with the membership function of the small category created in advance as described above.

For example, the membership functions of the small categories A, B, and C are shown in FIGS. 12, 13 and 14, respectively. At this time, the output of the feature parameter of the unknown extraction pattern is the reference pattern It is assumed that the output is s point for 1 and t point for reference pattern 2. Table 1
Is a graph showing how much the s point and the t point have a membership value with respect to the membership function of each pattern (how much they belong to each small category).

[0114]

[Table 1]

From the membership values shown in Table 1, AND
When the calculation is performed, the minimum value of the membership value of each extracted pattern is obtained. The membership value for the unknown pattern is 1 in the small category A, 0 in the small category B, and 0 in the small category C. Therefore, it can be determined that the unknown extraction pattern belongs to the small category A.

Further, the arithmetic mean of the membership values shown in Table 1 can be used as a judgment material. That is, in this case,
Since the small categories A, B, and C are 1, 0.5, and 0.2 respectively for the unknown extraction pattern, the degree of each pattern can be estimated. However, in the case of using arithmetic averaging, an answer may be obtained even if there is no feature value of a certain characteristic parameter. It is meaningful as an auxiliary operation when it is difficult. That is, the state significantly deviates from the state when the average value is obtained for some reason.
Membership value is 0 only with the components of two feature parameters
And if the other reference pattern is 1, AND
Arithmetic cannot give the correct answer. In this case, the second candidate can be used as an arithmetic mean.

In the sixth embodiment, the membership values are trapezoidal. However, any value may be used as long as it is a so-called convex fuzzy set having no depression in the membership function. Such an arithmetic device can be configured using a Neumann computer or using an arithmetic device using hardware.

Embodiment 7 In this embodiment 7, up to the point where the characteristic amount is obtained by using the characteristic amount detecting means such as an optical correlation arithmetic unit,
Since it is similar to the above-described first to third embodiments, only the part of the identification determination device 31 will be described.

The determination by the identification determination device 31 can be logically performed as described in the first embodiment depending on the distribution of the small categories, but all the characteristic parameters can be taken in a plurality of small categories. When the ranges have a stochastic overlap, it is effective to make a determination using a hierarchical neural network represented by a single-layer or multilayer perceptron. FIG. 15 shows an identification determining device 31 using a hierarchical neural network.
1 shows a configuration diagram of one embodiment.

In such a hierarchical neural network, a feature amount for each feature parameter of an extracted pattern belonging to each small category is input to each node of the input layer, and a desired output indicating the small category at that time is presented. By performing on all the extraction patterns a plurality of times, the connections in the neural network are organized so that the neural network outputs a small category suitable for the input extraction pattern. A category that cannot be completely separated by itself alone can also be determined. The feature amount of each feature parameter of the extracted pattern detected by the feature amount detection unit is input to each node of the input layer, and the unknown pattern is identified based on the output state of the node of the output layer.

The identification determining device 31 includes the computer 18
The determination function may be realized by software using, or the same function may be realized by hardware by an electronic circuit. There is no particular limitation on the learning algorithm. For example, in the case of a multi-layer perceptron structure having three or more layers, an error back propagation method based on a well-known generalized delta rule can be used.

The neural network also includes, for example, a network in which a small category is created by a hypersphere in a feature space, and a network in which a distance is weighted from a reference data point. There are functions and the like, and these may be used.

[0123]

According to the configuration of the optical pattern recognition / classification apparatus of the present invention as described above, it is not necessary to use a matched filter which draws interference fringes such as holograms which must be produced for each pattern to be extracted. In addition, the alignment of the optical system is relatively easy. Also, since the entire image is processed optically,
Scanning of a pattern to be extracted is not necessary, and therefore, pattern extraction and recognition classification can be performed in a short time. Furthermore, not only patterns that are exactly the same as the patterns to be extracted, but also patterns that have a certain degree of correlation can be easily and simultaneously extracted using a simple filter, and then finer classification can be performed. A pattern can be extracted and recognized and classified from an image such as a natural image including a pattern of many objects and characters.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram showing a configuration of an embodiment of an optical pattern recognition / classification device of the present invention.

FIG. 2 is a schematic configuration diagram of an optical system using a conventional matched filtering method.

FIG. 3 is an explanatory diagram showing an example of an input image according to the present invention.

FIG. 4 is an explanatory diagram showing an example of a representative pattern according to the present invention.

FIG. 5 is an explanatory diagram illustrating an example of an output image according to the present invention.

FIG. 6 is an explanatory diagram showing an example of an extraction pattern according to the present invention.

FIG. 7 is a schematic configuration diagram showing the configuration of another embodiment of the optical pattern recognition / classification device of the present invention.

FIG. 8 is a schematic configuration diagram showing the configuration of another embodiment of the optical pattern recognition / classification device of the present invention.

FIG. 9 is a schematic configuration diagram showing a partial configuration of another embodiment of the optical pattern recognition / classification device of the present invention.

FIG. 10 is a schematic configuration diagram showing a configuration of a part of the device of FIG. 9;

FIG. 11 is a schematic configuration diagram showing a partial configuration of another embodiment of the optical pattern recognition / classification device of the present invention.

FIG. 12 is an explanatory diagram showing a membership function created based on an output of a feature parameter of an extraction pattern of one small category according to the present invention.

FIG. 13 is an explanatory diagram showing a membership function created based on the output of the characteristic parameter of the extraction pattern of another small category according to the present invention.

FIG. 14 is an explanatory diagram showing a membership function created based on an output of a feature parameter of an extraction pattern of still another small category according to the present invention.

FIG. 15 is an explanatory diagram showing an example of a hierarchical neural network as an embodiment of the identification determination device of the optical pattern recognition / classification device of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Pattern extraction means 2 ... Optical correlation calculation means 3 ... Discrimination means 11 ... Laser 12,12a, 12b, 12c, 12d, 72,76,
80, 84 luminous flux 13 beam expander 14, 14a, 14b, 14c, 14d, 95 half mirror 15, 25, 53, 58 image display means 16, 20, 26, 29, 54, 59 Fourier transform lens 17, 27 spatial light modulator 18 computer 19, 28 Fourier transform image output means 21, 55, 60 screen 22, 56, 61 imaging means 23, 62 image processing device 24, 91, 96 mirror 30 ... Light receiving means 31... Identification discriminating device 41. Matched filters 51 and 52... Fourier transform lens array 63. Light receiving means array 71, 75, 79, 83, 101. 74, 82, 93 ... optical writing type spatial light modulator 78, 86, 94 ... polarizing beam splitter 90 ... Force image 92, 97 ... imaging lens 103 ... Dove prism

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-63-78283 (JP, A) JP-A-2-293827 (JP, A) JP-A-3-105569 (JP, A) JP-A-4- 84374 (JP, A) JP-A-5-100278 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06T ⁷ /00-7/60 G02B 27/46 G02F 3/00

Claims (9)

(57) [Claims] 1. A pattern belonging to a desired large category including a set of one or more small categories including one or more patterns to be recognized and classified is extracted from an input image (hereinafter, referred to as an input image). A pattern extracting unit, and an amount (hereinafter, referred to as a feature amount) representing each characteristic of the pattern extracted by the pattern extracting unit (hereinafter, referred to as an extracted pattern) is detected using an optical system that performs correlation or an operation similar thereto. The pattern extracting means, wherein the pattern extracting means comprises: a characteristic amount detecting means for determining the small category to which the extracted pattern belongs, based on one or more characteristic amounts detected by the characteristic amount detecting means. A light source that emits a first coherent light beam, and an input image displayed as an intensity transmittance or a reflectance distribution. A first image display unit for giving an intensity distribution to the luminous flux, a first Fourier transform lens for optically Fourier-transforming the intensity distribution of the light beam transmitted or reflected by the first image display unit; Patterns having characteristics common to all patterns included in the large category (hereinafter referred to as representative patterns)
First Fourier transform image output means for outputting a Fourier transform image of the following, and a Fourier transform image of the representative pattern, which is arranged on a Fourier transform surface of the first Fourier transform lens, converts the Fourier transform image of the representative pattern into an intensity transmittance or a reflectance distribution. A first spatial light modulator that further provides an intensity distribution to the first light beam; and optically further Fourier transforms the intensity distribution of the light beam transmitted or reflected through the first spatial light modulator. A second Fourier transform lens, an imaging unit that captures an output image obtained by Fourier transform by the second Fourier transform lens, and the output image as an input image of the first image display unit again. Image feedback means for displaying, and a position detection means for detecting a pixel having an output value equal to or more than a certain value in the output image, and detecting a position of a group of a certain number or more of the pixels. An area cutout unit that cuts out an area of an appropriate size from the input image based on the output from the position detection unit, and an image region cut out by the area cutout unit is A light source that emits a second coherent light beam, and a light source that emits a second coherent light beam, and one or more reference patterns are represented by an intensity transmittance or a reflectance distribution. A second image display unit for displaying and providing an intensity distribution to the second coherent light beam; and a third image optically Fourier-transforming the intensity distribution of the light beam transmitted or reflected by the second image display device. And a second Fourier transform image output means for outputting a Fourier transform image of the extracted pattern obtained by the pattern extracting means. A fourth Fourier transform lens disposed on the Fourier transform surface of the third Fourier transform lens, and displaying a Fourier transform image of the extracted pattern as an intensity transmittance or a reflectance distribution to further provide an intensity distribution to the second light flux; A second spatial light modulator, a fourth Fourier transform lens that optically further Fourier-transforms the intensity distribution of the light beam transmitted or reflected by the second spatial light modulator, and a fourth Fourier transform lens. Light receiving means for detecting the amount of light of the image obtained by Fourier transform in accordance with the one or more reference patterns, and wherein the amount of light of the image detected by the light receiving means is An optical pattern recognition / classification apparatus, which corresponds to a feature amount detected by the feature amount detection means. 2. A pattern belonging to a desired large category comprising a set of one or more small categories including one or more patterns to be recognized and classified is extracted from an input image (hereinafter referred to as an input image). A pattern extracting unit, and an amount (hereinafter, referred to as a feature amount) representing each characteristic of the pattern extracted by the pattern extracting unit (hereinafter, referred to as an extracted pattern) is detected using an optical system that performs correlation or an operation similar thereto. A feature amount detecting unit that performs the determination on the small category to which the extracted pattern belongs based on one or more feature amounts detected by the feature amount detecting unit. A light source that emits a first coherent light beam; and an input image displayed as an intensity transmittance or a reflectance distribution. A first image display means for giving an intensity distribution to the light beam; a first Fourier transform lens for optically Fourier transforming the intensity distribution of the light beam transmitted or reflected by the first image display means; Patterns that have characteristics common to all patterns included in the large category (hereinafter referred to as representative patterns)
First Fourier transform image output means for outputting a Fourier transform image of the following; and a Fourier transform image of the representative pattern, which is arranged on a Fourier transform surface of the first Fourier transform lens, is converted into an intensity transmittance or a reflectance distribution. A first spatial light modulator for giving an intensity distribution to the first light beam, and further optically Fourier transforming the intensity distribution of the light beam transmitted or reflected by the first spatial light modulator. A second Fourier transform lens, imaging means for taking an output image obtained by Fourier transform by the second Fourier transform lens, and displaying the output image again as an input image of the first image display means In the output image, in the output image, a pixel having an output value of a certain value or more is detected, and a position detecting means for detecting a position of a group of a certain number or more of the pixels. Based on the output from the position detecting means, an area extracting means for extracting an area of an appropriate size from the input image, and the image area extracted by the area extracting means is the pattern A light source that emits a second coherent light beam, and one or more reference patterns are represented by an intensity transmittance or a reflectance distribution. A second image display means for displaying and giving an intensity distribution to the second coherent light flux, wherein each lens is arranged corresponding to the position of the one or more reference patterns, and A first Fourier transform lens array for optically Fourier transforming the intensity distribution of the light flux transmitted or reflected by the means for each reference pattern; A second Fourier transform image output means for outputting the Fourier transform image of the extraction pattern obtained by extracting means, the image of the Fourier transform image of the extracted pattern, the first
An image processing device that is arranged in parallel at a position corresponding to each lens of the Fourier transform lens array; and the image processing device that is disposed on the Fourier transform surface of each lens of the first Fourier transform lens array, and is created by the image processing device. A third spatial light modulator that displays a parallel image of the Fourier transform image of the extracted pattern as an intensity transmittance or a reflectance distribution to further provide an intensity distribution to the second light flux; A second Fourier transform lens array arranged corresponding to the positions of the plurality of reference patterns and optically further Fourier-transforming the intensity distribution of the light flux transmitted or reflected by the third spatial light modulator; The amount of light of the third output image obtained by performing Fourier transform by each lens of the two Fourier transform lens arrays corresponds to the one or more reference patterns. An optical pattern recognition classification comprising: a light receiving means for detecting; and a light amount of an image detected by the light receiving means corresponding to a feature amount detected by the feature amount detecting means. apparatus. 3. A method according to claim 1, wherein at least one of said first and second Fourier-transformed image output means comprises: a light source for emitting a coherent light beam; A third image display means for displaying and providing an intensity distribution to the light beam; a fifth Fourier transform lens for optically Fourier transforming the intensity distribution of the light beam transmitted or reflected from the third image display means; 3. An imaging unit configured to capture a Fourier transform image of the representative pattern or the extracted pattern obtained by the fifth Fourier transform lens. 4.
An optical pattern recognition / classification device according to claim 1. 4. The spatial light modulator according to claim 1, wherein at least one of said first, second and third spatial light modulators is a first optical writing type spatial light modulator. The optical pattern recognition / classification apparatus according to any one of the preceding claims. 5. The spatial light modulator of claim 1, wherein the first image display means forms an input image on a writing surface of the spatial light modulator of the second optical writing type. The optical pattern recognition / classification device according to any one of claims 1 to 4, wherein the optical pattern recognition / classification device comprises a system. 6. A Fourier transform image of the representative pattern or the extracted pattern is provided in at least one of the first, second, or third spatial light modulators by an appropriate threshold substantially in real time. The optical pattern recognition and classification device according to any one of claims 1 to 3, wherein the optical pattern recognition and classification device is displayed as a binarized image. 7. At least one of the first and second optical writing type spatial light modulators performs an operation of binarizing an image written from a writing surface with a threshold and displaying the image. The optical pattern recognition / classification device according to claim 4 or 5, wherein: 8. The discriminating means, for each small category,
The feature amount of the extracted pattern belonging to this small category is acquired a plurality of times, and based on the representative value of the obtained set of outputs,
A membership function of the extracted pattern is created in advance for each component (hereinafter, referred to as a feature parameter) of the extracted pattern and the feature amount, and the feature amount of the extracted pattern extracted from the input pattern is used for each of the small categories. A membership value is calculated for each feature parameter based on each membership function, and an average value or a minimum value of the obtained membership values is set to a degree belonging to each of the small categories of the extraction pattern. The optical pattern recognition / classification apparatus according to any one of claims 1 to 7, further comprising an arithmetic unit for determining an extraction pattern. 9. The discrimination means is a neural network which inputs each feature amount acquired for each of the extraction patterns and has an output indicating a small category of the extraction pattern, and discriminates the extraction pattern from the output. The optical pattern recognition / classification device according to any one of claims 1 to 7, wherein: