JPH0756122A - Image processor for feature extraction - Google Patents

Abstract

PURPOSE:To extract features from an image of a body which moves extremely fast or three-dimensional body while it is irradiated with natural light by reading light-dark information on write light corresponding to whether or not the write light is present or its intensity and performing a 2-input, 1-output AND operation through a read-side optical system of orthogonal Nicol constitution. CONSTITUTION:An image is inputted while its light intensity is adjusted and the incoherent-coherent transformation of the input image is performed by a 1st polarized light modulating means 210; and a 2nd polarized optical modulating means 240 ANDs the image which is Fourier-transformed by a Fourier transforming means 220 with a mask image which is inputted through a mask input means 230 to extract the features from the image, and a series of processes are optically performed until the image is formed and outputted by an image output means 260, thereby performing real-time feature extraction. Therefore, even when the image to be processed and the mask image are irradiated with the natural light, the image processes which are fast and superior in operability can be performed.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus for feature extraction, and more particularly to extracting feature information such as a contour of an input image or a line segment in a specific direction and improving the quality of the input image by mask processing. The present invention relates to an image processing device. More specifically, the two-dimensional parallel optical operation is performed on the optical input image as it is without any electrical processing means, and is particularly useful in the field of optical image processing such as image recognition applications. It is something.

[0002]

2. Description of the Related Art Conventionally, an electric method, an optical method, and a method using a special element are known as methods for extracting image features. Here, in particular, the conventional technique will be described by considering the feature of extracting a directional line segment. For example, the extraction of the contour information of the image and the specific frequency band information or the improvement of the quality of the input image will be similarly described. can do. Here, the directional line segment means a line element in a specific direction (for example, the vertical direction or the horizontal direction) among the line segments forming the outline of each object in the image.

An electrical method will be described as the first conventional technique. For example, in an electrical method in which an image to be processed is input using an image pickup means such as a CCD camera and an optical signal from the input image is once converted into an electrical signal,
An image at a certain time is stored in an image memory, and a mask process or a raster calculation is executed to extract a directional line segment.
In this method, the frame rate required to store an electrical signal in the frame memory, for example, 1 / screen
Calculation time of 30 seconds or more is required.

In the electrical method, parallel processing is generally costly even if the spatially parallel image information at the same time, which is originally spatially spread, is partially utilized in parallel (for example, a parallel arithmetic processor). However, it is often the case that image information is expanded in time and processed serially. Therefore, the application range may be limited depending on the processing time.

On the other hand, a high-speed input method has been proposed in which the frame rate of the input device is faster than that of the above-mentioned image pickup means, that is, a moving image of an object moving at high speed is continuously input using a high-speed camera or the like. According to this, normally, the input image is once recorded in a storage medium such as an image memory or a recording tape, and after the recording is completed, the feature extraction processing is performed for each recorded frame.

Next, the accuracy of the extraction of the directional line segment by the electric mask processing will be described. In an electrical mask process for extracting a directional line segment of an image recorded in an image memory or the like, a mask pattern corresponding to the directional line segment to be extracted is usually formed. In the case of the electrical method, since the two-dimensional spatial position information of the image corresponds to the memory address of the image memory, a quantization error relating to the position information occurs when converting the spatial position information into the memory address. .

Further, in the case of extracting a directional line segment included in a direction of a certain range, not limited to a specific direction (for example, in the case of extracting a line segment group forming an angle of 0 to 15 degrees with respect to the horizontal direction, etc.). In (), a plurality of types of mask patterns having different directivities within the range to be detected are produced, an operation for extracting a directional line segment is performed for each mask pattern, and the logical sum of the respective results is output. Therefore, when extracting a directional line segment in a certain range, the number of processing steps increases.

An optical method will be described as the second conventional technique. It is known that the high speed property of light can be utilized in an optical method for extracting image features by filtering processing in the frequency domain by optical Fourier transform using a lens. On the other hand, in the optical method, it is known that the field of application is limited by the characteristics of the filter and the input image. The optical method is described in, for example, "Applied Optics I", by Tatsuro Suzuki, published by Asakura Shoten. The features of the optical method will be briefly described below.

The production of a filter requires a fine and delicate process such as recording a filter pattern on a photographic film or vapor deposition on a glass plate, and the material of the filter is reusable such as a film or a glass plate. Is a low material.

Furthermore, the spatial filtering method in the frequency plane exhibits desirable performance when extracting or removing a feature (for example, a line drawing) having a substantially constant spatial frequency contained in the input image. However, in a more general feature extraction process, various spatial frequencies included in an image are mixed from low to high, and a desired frequency range of features to be extracted may be wide. Therefore, when one kind of fixed filter in the frequency plane is applied to feature extraction having a wide range of spatial frequencies, only nonuniform and unclear features are obtained for the entire image.

The input image, which can be processed by optical Fourier transform and filtering, is an image carried by light of a single wavelength (usually exhibiting coherent light). Therefore, in order to perform the optical filtering process, coherent light representing the characteristics of the object is formed by some irradiation means and used as the input image. For example, when processing an image under natural light irradiation conditions, the image is usually photographed on a photographic film, and single-wavelength light such as laser light is irradiated from the back surface to form an input image, or At the laboratory level, once the wavelengths are aligned via the incoherent-coherent conversion element, the input image is obtained. When photographing on a photographic film, a development process is necessary, and even when an incoherent-coherent conversion element is used, the response time of the element is delayed.

Up to now, there has been reported an example of drawing a Lissajous figure on a BSO element with a laser beam and using it as a filter (Applied Optics Vol.15, No.6, 1976).
However, there is a problem from a practical point of view in terms of the method of controlling the BSO element, the energy to be applied and the response speed. Similarly, in the same report, the BSO element is also used as an incoherent-coherent conversion element, but a two-dimensional photographic film is used as an input image from the point of the sensitivity of the element, and a coherent light image is obtained by irradiating laser light from the back surface It remains at the experimental level to convert to. Therefore, it is difficult to apply to real-time real objects.

As a third prior art, a special optical element will be described. For example, “Optical Neurochips for Direc.
t Image Processing (optical neurochip for direct image processing) ", Kuman et al., in Proceedings D-83 of the 1992 IEICE Spring Conference. Number of elements pixels (32 x 32)
Since there are few, the accuracy of feature extraction is restricted. Further, the application range may be limited depending on the control method of the element, the amount of energy to be applied, and the response time of the element.

[0014]

In the field of image processing [SUMMARY OF THE INVENTION], 2-dimensional, (representing translation and rotation in the plane) 2 _^1/2-D, or not only the image processing of the still object 3,
Image processing of a three-dimensional object or an object that moves at high speed is required. Further, since the image processing apparatus itself is installed on a moving object or used in various environments, there is a strong demand for downsizing the apparatus, reducing costs, and simplifying operation and maintenance. In response to these demands, the above-mentioned conventional techniques include the following problems.

In the electrical feature extraction method, the feature extraction processing speed is slow because the input circuit and the electric circuit whose response speed is slower than the speed of the optical signal are interposed.

For example, as described above, since the calculation time depends on the frame rate of the input device, it cannot be captured within the frame rate, that is, the feature of the object moving at a very high speed as compared with the frame rate, for example, the direction. The sex segment cannot be extracted in real time. On the other hand, in the case of input from a high-speed camera or the like, since the image is temporarily stored, the real-time property of the feature extraction processing cannot be expected at all.

A spatial specific direction capable of reducing a quantization error when converting a spatial position into an image memory address, for example, a vertical direction or a horizontal direction, or 4 for them.
Compared with the accuracy of the extraction of the directional line segment in the 5 degree direction, the extraction accuracy for the other directional line segments is poor. Here, the mask size is increased or the mask type is increased in order to increase the directional extraction accuracy. If the number is increased, the processing time becomes longer.

Further, when processing an image containing a fine structure, it is necessary to increase the number of pixels of the input image, which increases the processing time.

In the case of the optical feature extraction method, since the filter pattern is manufactured by using a photographic film or a glass plate, it is inevitable that the filter is difficult to manufacture.
The filter replacement requires mechanical replacement work, and it is difficult to replace the filter in real time according to the application.

Furthermore, the filter material such as film and glass plate cannot be reused, which increases the manufacturing cost of the device. In addition, since unnecessary phase modulation may be added to a photographic film or the like, the film needs to be immersed in liquid to alleviate it, which complicates the entire apparatus.

Further, the optical method cannot deal with all the spatial frequencies contained in the input image with one kind of fixed filter. For example, since the spatial frequency components included in the processing target image are different depending on whether the size of the processing target image is large or small, the size of the processing target image may be reduced when performing feature extraction using a circular filter pattern. It is necessary to change the radius of the circular filter pattern according to.

When the image processing is performed by the optical Fourier transform, the image to be processed is limited to the image carried by the light of a single wavelength. Therefore, the natural light image cannot be handled as it is as an input image.

In the method using a special optical element, the number of pixels is limited, the element control method is complicated, and the energy to be applied is large, and moreover, real-time image input and feature extraction processing such as moving image processing are required. There is a problem that the response speed of the device is slow for the application requiring the.

In order to solve the above problems of the prior art,
Improved feature extraction processing performance by speeding up processing, inputting images under natural light irradiation conditions, and reducing restrictions on feature extraction processing due to the spatial frequency included in the input image. A device design that realizes cost reduction is required.

In view of the above points, the present invention provides an object that moves at a very high speed under natural light irradiation within a time period below the frame rate of a conventional electric image pickup device by two-dimensional parallel light calculation. Another object of the present invention is to provide a feature extraction image processing apparatus that performs feature extraction processing of an image of a three-dimensional object, performs feature extraction of directional line segments included in an input image, and performs mask processing such as image quality improvement. To do.

The present invention further comprises a low-cost image processing apparatus for feature extraction, which is simply constituted by only an optical element and optical components which consume little energy and optical signal loss, consume less energy in the apparatus, and can be easily controlled. The purpose is to provide.

[0027]

FIG. 1 is a block diagram showing the principle of the present invention.

The image processing apparatus for feature extraction according to the present invention is
Polarization modulation means for receiving the reading light and the writing light and changing the polarization angle of the reading light according to the light intensity of the writing light, and means for supplying the light of the mask image used for the mask processing as the writing light to the polarization modulating means. And a means for optical Fourier transforming the light of the image to be processed into the read light to the polarization modulation means, and a light that is polarized at a predetermined angle out of the output light of the polarization modulation means is masked. It is composed of a processed image output means for taking out light and performing optical Fourier transform of the light for output.

The image processing apparatus for feature extraction according to the present invention is
A first polarization modulator that receives the read light and the write light and changes the polarization angle of the read light according to the light intensity of the write light, and receives the read light and the write light and receives the read light and the write light according to the light intensity of the write light. Second polarization modulating means for changing the polarization angle of the reading light, means for making the light of the image to be processed into writing light for the first polarization modulating means, and optical Fourier transform for the output light of the first polarization modulating means. Means for making the read light to the second polarization modulating means, means for supplying the light of the mask image used for the mask processing as write light to the second polarization modulating means, and the output of the second polarization modulating means. The operation of the processed image output means for passing the light polarized at a predetermined angle out of the light, extracting the masked light, and performing the optical Fourier transform of this, and the operation of the first and second polarization modulating means Control It becomes more and your means.

[0030]

The polarization modulation means of the image processing apparatus of the present invention outputs the light / dark information of the writing light as a difference in the polarization state of the reading light in accordance with the presence or absence of the writing light, and the intensity of the writing light. The feature extraction mask processing is executed by a logical product operation of two inputs and one output having an orthogonal Nicols configuration. In other words, the polarization modulation means uses an image obtained by Fourier transforming the input image carried by the light of a single wavelength and linearly polarized light in real time by the lens as the readout light of the polarization modulation means,
On the other hand, the mask image of the mask pattern formed in real time by another lens under irradiation of natural light is written as writing light in the polarization modulator, and an optical signal corresponding to the logical product of the reading light and the writing light is output. The output signal is Fourier transformed by the image output means and imaged on the output surface.

As a result, the processing is performed optically, so that high speed and energy saving can be obtained. In addition, a general two-dimensional image can be input as a mask image without producing a special mask pattern, the mask can be easily replaced according to the purpose, and various types of features of the image are extracted in time series.

The feature extraction image processing apparatus according to the present invention is
An image is input while adjusting the light intensity, an incoherent-coherent conversion of the input image is performed by the first polarization modulation unit, an image Fourier-transformed by the Fourier transform unit, and a mask image input by the mask input unit. Is performed by the second polarization modulation means to extract the characteristics of the image, and the series of processes are optically performed until the image is output by the image output means. Specific characteristics are extracted. Therefore, even when the image to be processed and the mask image are under irradiation of natural light, image processing with high operability can be performed at high speed.

In the image processing apparatus for feature extraction of the present invention, the image to be processed is Fourier-transformed,
When the feature extraction mask processing is performed using the mask pattern that blocks the DC component (Oth order light) of the image, the contour image in the processing target image is output. When the feature extraction mask processing is performed using a mask pattern that blocks directional components other than the specific direction of the image, an image in which a line segment in a direction perpendicular to the specific direction in the processing target image is extracted is output. When the feature extraction mask processing is performed using a mask pattern that blocks a part of the DC component of the image, an image in which the high-frequency component in the processing target image is emphasized, that is, an image in which the contour portion is emphasized is output. When the feature extraction mask processing is performed using a mask pattern that blocks higher-order components of the image, an image in which the low-frequency components in the processing target image are emphasized, that is, an image in which the contour portion is smoothed is output. When the feature extraction mask processing is performed using a mask pattern that blocks components other than the specific order component of the image, an image in which the specific frequency component in the processing target image is emphasized is output. For example, as described above, various types of mask processing are realized.

[0034]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail below with reference to the drawings. This embodiment will be described in the order of system configuration, two-dimensional optical element, mask processing, image input for processing, incoherent-coherent conversion, optical Fourier transform, mask image input, control section, and image output section. Further, a feature extraction process according to an embodiment of the present invention for a processing target image including substantially uniform spatial frequencies and a processing target image including various spatial frequencies will be described.

FIG. 2 is a diagram showing the system configuration of an embodiment of the present invention. The figure shows a processing target image 10 that is an input to the system, a mask pattern 30 used for the characteristic extraction processing of the processing target image 10, a characteristic extraction result 40 that is the result of the system characteristic extraction processing, the processing target image 10 and the mask. The image forming lens 50 and the image forming lens 51 that input the pattern 30 respectively, the light source 80, and the two-dimensional optical element 20 that configures the incoherent-coherent conversion unit together with the polarization beam splitter 70, and the optical Fourier transform unit of the processing target image 10 A certain Fourier transform lens 60, a two-dimensional optical element 21 and a polarization beam splitter 71 forming a logical product operation section of mask processing, a two-dimensional optical element 20, a two-dimensional optical element 21, and a control section 100 for controlling the operation timing of the light source 80. ,
Fourier transform lens 61 for outputting the result of the feature extraction processing
Indicates.

The image 10 to be processed is converted into a two-dimensional signal 82 modulated by the reading light 81, further becomes a coherent light image 83 via the polarization beam splitter 70, and a Fourier transform image 84 of the coherent light image is obtained by the Fourier transform lens 60. And becomes the reading light of the two-dimensional optical element 21. On the other hand, the mask image real image 31 of the mask pattern 30 formed by the lens 51 is the two-dimensional optical element 2
The writing light becomes 1. A signal 85, which is modulated by the writing light 31 and is extracted and which represents the characteristic of the processing target, is polarized by the polarization beam splitter 71 and output as a logical product 86 of the writing light 31 and the reading light 84. The logical product 86 is Fourier transformed by the Fourier transform lens 61 and output to the feature extraction result 40.

Next, regarding the two-dimensional optical element used for incoherent-coherent conversion and AND operation in the image processing apparatus for feature extraction of the present invention,
As a typical two-dimensional optical element, for example, a paper by Fukushima et al. “Bistable spatial light modulator using ferroelectric liquid crystal (“ Bista
ble spatial light modulator using a ferroelectric
liquid crystal ")", Optics Letters. Volume 15, Issue 285 (1
The ferroelectric liquid crystal spatial light modulator described in 990) will be described. This element is a reflective and photo-addressable two-dimensional optical element using liquid crystal. It is also possible to use a transmissive and photo-addressable two-dimensional optical element using liquid crystal. In this case, an optical system for image input, a light source and an element so that the writing light and the reading light are incident from the same direction. And a polarizing plate may be used instead of the polarizing beam splitter.

This element converts the light intensity distribution state of the input image into a polarization distribution state and outputs it as two-dimensional light information. For the purpose of feature extraction, it is possible to use another element without being limited to this liquid crystal element. However, in practical use, the element should be selected in consideration of the response speed, sensitivity and energy consumption of the element. The ferroelectric liquid crystal spatial light modulator used in one embodiment of the present invention has a high sensitivity to brightness and a fast response speed, so that an image of a real object illuminated by natural light is used as writing light and reading light for the element. Available. Further, by using the coherent light as the reading light, it is possible to perform the incoherent-coherent conversion of the input image which is the writing light of the element.

In the image processing apparatus for feature extraction of the present invention, one incoherent-coherent transform, two Fourier transforms of the input image, and the logical product operation of the mask pattern are used, respectively, for a total of two bodies.

FIG. 3 is a diagram for explaining the structure of the two-dimensional optical element. In FIG. 3, (a) is a transparent electrode (I
Between two glass coated with (TO), a photoconductive film (a-Si: HPC) of hydrogenated amorphous silicon, a dielectric mirror (DM), and a ferroelectric liquid crystal (FLC).
The structure of the element sandwiched between is shown, and the thickness of each layer is 1
It is about μm. In FIG. 3A, erasing light and writing light as an input are emitted from the left side. On the other hand, the reading light is emitted from the right side, and the result of the reflection is output. Note that this element usually has a writing sensitivity in the visible light region, but if the photoconductive film has appropriate characteristics, it can write an optical image in the infrared region or the ultraviolet region.

FIG. 3B shows the rubbing direction of the transparent electrode (ITO) for the purpose of stabilizing the initial state of the liquid crystal as viewed from the reading side. The element is preferably a rectangle suitable for optical image processing applications and is rubbed at an angle of 22.5 degrees to the sides of the rectangle for maximum use of effective area. The selection of the rubbing angle will be described later.

FIG. 4 is a diagram for explaining an embodiment of controlling the operation timing of the two-dimensional optical element. In the figure, the horizontal axis represents time, the vertical axis represents voltage, and E, W, and R represent the timings of erasing, writing, and reading, respectively. FIG. 4A shows an electric control pulse given to drive the two-dimensional optical element, while FIG. 4B shows a timing at which the reading light is emitted.

At timing E, the erasing light is turned on and the element is refreshed over the entire surface. At timing W, writing to the element is performed, and at timing R, the reading light is turned on and the two-dimensional light of the element is turned on. The information is read. The entire pulse with EW-R as one cycle repeats. In order to generate the read pulse R, for example, there is a method of optical pulse modulation of read light using an acoustooptic device or a method of directly modulating a semiconductor laser as a light source. As the erasing light, the entire surface of the two-dimensional optical element may be irradiated with an LED or the like.

When the input image at the time t is written in the two-dimensional optical element, the image is held in the element for the time R. Therefore, one cycle of E-W-R is set to a time shorter than the frame rate (for example, 1/30 seconds) of the electrical input means, and the width of W is made smaller so that the actual rate is equal to or less than the frame rate. A moving image that changes with time can be used as the processing target input image. As control according to one embodiment of the present invention, applied voltage V = ± 10 V, E = 300 μ
A two-dimensional optical element can be operated with s, W = 200 μs, and R = 500 μs, and one cycle in this case is 1 ms.
Becomes Further, the LED as the erasing light and the acousto-optic element used for the optical pulse modulation of the reading light can be driven at about 5V, respectively.

It is possible to set one EW-R cycle to a time longer than the electrical frame rate, if necessary.

FIG. 5 is a diagram for explaining the operating principle and manufacturing conditions of the two-dimensional optical element. (A) of the same figure shows the orientation angle of the liquid crystal molecules as seen from the read side of the device, and the read light is assumed to be incident from the upper side of the paper surface and reflected toward the upper side again.

First, in the input image, the liquid crystal is oriented in the ON (+) direction with respect to the place where the writing light is (bright), and conversely, to the place where there is no writing light (dark). The liquid crystal is aligned in the OFF (-) direction. θ _ON , θ _OFF
Are tilt angles of the liquid crystal that are counterclockwise or clockwise with respect to the rubbing direction. The read light is modulated by the effect of the rotation (tilt) of the liquid crystal.

Explaining the relationship between the input image and the rotation of the liquid crystal in more detail, the erasing timing E shown in FIG.
In the above, the liquid crystal on the entire surface of the device is rotated by θ _OFF degrees in the OFF direction by the irradiation of the erasing light and the application of the pulse to the device and is held as it is. At the next writing timing W, by applying a writing pulse to the element, only the liquid crystal in the portion where the writing light is present (bright) is rotated by θ _ON degrees in the ON direction and held. Therefore, at the read timing,
Two states, the OFF direction and the ON direction, are read.

In this embodiment, θ _ON is described as counterclockwise rotation and θ _OFF is described as clockwise rotation, but they may be reversed. In fact, when the positive and negative pulses of the pulse shown in FIG. 4A are applied to this element by reversing, the reading light is modulated by the rotation of θ _ON clockwise and θ _OFF counterclockwise.

By the way, in the case of the reflection type element, the read light enters the element from the read surface side shown in FIG. 3A, is reflected by the dielectric mirror (DM), and is output. Pass through the liquid crystal layer twice. Therefore, the read light is modulated at an angle twice the tilt angle of the liquid crystal. FIG. 5B shows the polarization direction of the light reflected and output by the element. On the other hand, in the case of the transmissive element, the liquid crystal layer passes through only once, and the modulated angle matches the tilt angle.

In FIG. 5B, the read light is linearly polarized in the horizontal axis direction shown by a polarizer, and the read light is orthogonal to the analyzer (Anal).
yzer) is the linearly polarized light of the vertical component only.
That is, the optical system on the reading side is set to the crossed Nicols state. The orthogonal Nicols state can be configured with a polarizing beam splitter. When a transmissive element is used, two polarizing plates are used to form a crossed Nicols.

It is conceivable to set the optical system on the reading side to the parallel Nicol state, but in this case, a bright / dark inverted image of the output in the orthogonal Nicol state is output. It should be noted that when a reading optical system other than the crossed Nicols and parallel Nicols states is set, the contrast of the output image becomes low.

In the orthogonal Nicol read-out optical system, the read-out light intensities I _PS ^ON and I _PS ^OFF reflected by the elements are expressed by the equations (1) and (2). I _PS ^ON shows the light intensity when there is writing light, and I _PS ^OFF shows the light intensity when there is no writing light.

I _PS ^ON = sin ² {2 (θ + θ _ON )} ・ sin ² (2πdΔn / λ) (1) I _PS ^OFF = sin ² {2 (θ−θ _ON )} ・ sin ² (2πdΔn / λ) (2) where θ is the angle formed by the rubbing direction and the polarization direction of the readout light (hereinafter, θ is referred to as the mounting angle), θ _ON , θ
_OFF , with or without writing light,
Tilt angle of ferroelectric liquid crystal based on rubbing direction,
d is the thickness of the ferroelectric liquid crystal layer, Δn is the birefringence of the ferroelectric liquid crystal, and λ is the wavelength of the reading light. From equations (1) and (2), it can be seen that the period of I _PS is θ = 90 degrees.

In the case where an input image is written in this element and the operation of outputting the brightness information of the input image as it is, in order to obtain the best contrast and maximum reflectance of the element, θ _ON = θ _OFF = When a liquid crystal with an angle of 22.5 degrees is selected, when λ = 633 nm and Δn = 0.15, d = 1 μ
It is only necessary to prepare a device having m and to set θ = 22.5 degrees. This is clear from equations (1) and (2). Note that this θ = 22.5 degrees corresponds to the rubbing direction with respect to the sides of the rectangular element.

Next, the logical product operation by the two-dimensional optical element for the feature extraction mask processing will be described in detail based on an embodiment of the present invention.

FIG. 6 is a diagram for explaining a logical product by a two-dimensional optical element. FIG. 6A shows a two-dimensional optical element 20.
2 represents the bright and dark image written in, that is, the processing target image 10 in FIG. On the other hand, FIG. 6B shows the reading light given to the reading side, that is, the reading light 81 in FIG. 2, but in this case, the reading light is parallel light. As described above, in the crossed Nicols configuration, only the portion modulated by the writing light is output as the bright portion, so that the same bright / dark state result as the input as in FIG. 6C is output. As a result, the bright portion in FIGS. 6A and 6B has a logical value of 1
When the dark portion is made to correspond to the logical value 0, the output (c) of FIG. 6 represents the logical product of the images of (a) of FIG. 6 and (b) of FIG. In one embodiment according to the present invention, a logical product is calculated using the above method.

Next, the masking process of the target image by the logical product and the output of the processing result will be described in detail with reference to FIG. 2 according to an embodiment of the present invention.

The Fourier-transformed image in which the two-dimensional light information 83 is Fourier-transformed by the lens 60 is linearly polarized light and becomes the read light 84 of the two-dimensional optical element 21. On the other hand, the result that the mask pattern 30 is imaged on the two-dimensional optical element 21 by the lens 51 is the writing light 31 of the two-dimensional optical element 21. The read light 84 irradiated on the two-dimensional optical element 21 is
The polarization state is modulated according to the light intensity distribution of the mask pattern 30 and then reflected. Here, since the reading side optical system constitutes the orthogonal Nicols by the polarization beam splitter 71, the Fourier transform image 84 of the processing target image 10 and the mask are included in the two-dimensional optical signal 85 reflected by the two-dimensional optical element 21. An optical signal 86 corresponding to the logical product of the writing light 31 of the pattern 30 is output.

To briefly explain the function of the two-dimensional optical element used for the logical product processing, the writing light of the mask pattern is input and is input from the reading side in correspondence with the light intensity distribution of the mask pattern. The polarization direction of linearly polarized light is modulated, and the polarization distribution state is output as two-dimensional light information.
In one embodiment of the present invention, the ferroelectric liquid crystal spatial light modulator, which is the above-mentioned reflection type and photo-addressing type two-dimensional optical element, is incorporated in the feature extraction image processing apparatus.

Next, the input of the image to be processed will be described in detail with reference to FIG. The image 10 to be processed is formed as an image of an appropriate size on the two-dimensional optical element 20 by the image forming lens 50. The lens 50 has a zoom function and a focusing function, such as moving back and forth in the optical axis direction according to the size and distance of the image to be processed 10 to focus on the optical element 20. When a three-dimensional real object is used as the processing target image 10, there is also an advantage that the focusing point by the focusing function makes clear the light and dark information of the image. Further, the lens 50 may have a diaphragm function for adjusting the light amount of the input image.

The above-mentioned iris diaphragm is used as a means for adjusting the light intensity of the image to be processed. The iris diaphragm is conveniently incorporated in the lens 50, for example, and a normal camera lens can be used as an example.
With the diaphragm open, that is, in a state where a large amount of light is taken in, the lens 50 converts the image 10 to be processed into the two-dimensional optical element 20.
If the image is formed on the upper side, fine noise and the like in the image to be processed 10 can be removed from the characteristic. Further, when the aperture is slightly narrowed, a gradation image is written due to the characteristics of the two-dimensional optical element 20, the allowable range of brightness of the processing target image 10 is expanded, and the accuracy of the extracted features is monitored. However, the image processing apparatus of the present invention can be adjusted interactively, for example.

Further, as a means for adjusting the light intensity of the mask pattern, for example, an iris diaphragm is also incorporated in the lens 51, and its configuration and function are similar to those of the lens 50.

In the image processing apparatus of the present invention, Fourier transformation and spatial filtering are performed, and the two-dimensional optical element 2
The two-dimensional optical signal 83 serving as the readout light 84 of 1 must be an image carried in light of a single wavelength (monochromatic light).

First, when an image illuminated by natural light is used as an input image of the image processing apparatus of the present invention, the input image may be converted into an image carried by monochromatic light by incoherent-coherent conversion described later. . Therefore, the processing target image 10 is a real object illuminated by natural light or laser light, an image displayed on a display element such as a CRT display, and an image displayed on a spatial light modulator such as a liquid crystal panel and illuminated by light. Any image may be used as long as it has a light intensity distribution. On the other hand, if the image to be processed 10 is an image that is originally conveyed as monochromatic light, incoherent-coherent conversion is not necessary.

As described above, the two-dimensional optical signal 83 shown in FIG.
Is an image carried in light of a single wavelength (monochromatic light).
Further, since the two-dimensional optical signal 83 is the read light of the two-dimensional optical element 20, it must have an optimum wavelength depending on the manufacturing conditions of the element, and must be linearly polarized light.

First, when the light source 80 outputs random polarized light, linearly polarized light may be extracted as the reading light 81 by the polarization beam splitter 70. On the other hand, the light source 8
When 0 outputs linearly polarized light, the amount of the readout light 81 becomes maximum if the installation angle of the light source 80 is set so that the outputted linearly polarized light becomes S-polarized light. The linearly polarized light obtained by either of the above two methods becomes the reading light 81 of the two-dimensional optical element.

The wavelength of the light source 80 is, for example, a visible light semiconductor laser (wavelength 670 nm) or a He--Ne laser (wavelength 6).
33 nm) or the like can be used. Or light source 8
A wavelength selection filter may be installed after 0 to extract only a desired wavelength.

Further, in order to synchronize the irradiation timing of the reading light 81 with the reading timing of the two-dimensional optical element 20, for example, in the visible light semiconductor laser, the light source 80 may be directly modulated. On the other hand, when a continuous wave He—Ne laser is used, an acousto-optical element or the like may be installed after the light source to modulate light (presence or absence of light).

Next, the incoherent-coherent conversion for inputting an image illuminated by natural light to the image processing apparatus according to the present invention will be described. Image to be processed 1
0 is imaged on the writing side of the two-dimensional optical element 20 by the imaging lens 50 and written in the element. On the other hand, the linearly polarized readout light 81 whose traveling direction is bent by the polarization beam splitter 70 is applied to the readout side of the two-dimensional optical element 20,
Then it is reflected. The polarization state of the reflected two-dimensional optical signal 82 is modulated according to the light intensity distribution of the processing target image 10. Here, since the optical system on the read side constitutes the orthogonal Nicols by the polarization beam splitter 70,
Of the two-dimensional optical signal 82, the two-dimensional optical signal 83 obtained by converting the processing target image 10 into coherent light and linearly polarized light can be obtained.

Next, the structure of the crossed Nicols using the polarization beam splitter 70 according to the embodiment of the present invention will be described. The polarization beam splitter 70 has a property of transmitting P-polarized light, which is a linearly polarized light in the vertical direction, in a plane perpendicular to the traveling direction of light, of the polarized component of light, and reflecting S-polarized light orthogonal thereto and bending the traveling direction. . By utilizing this function, the read-out optical system of the two-dimensional optical element 20 can have a crossed Nicol configuration. That is, the purpose is to efficiently enter the read light 81 into the two-dimensional optical element 20, further efficiently output the read result of the two-dimensional optical element 20, and to reduce the space of the entire device.
In this case, it is necessary to use the optimum polarization beam splitter 70 corresponding to the wavelength of the reading light 81. In order to form the crossed Nicols, two polarizing plates may be combined instead of the polarization beam splitter. Similarly, the polarized beam splitter 71 has a reading optical system of the two-dimensional optical element 21 having an orthogonal Nicol configuration.

Next, a method of performing the Fourier transform of the image to be processed 10 on the two-dimensional optical element 21 according to the embodiment of the present invention will be described. The lens 60 Fourier transforms the coherent light image 83 on the read side of the optical element 21 as a Fourier transform image 84. Here, the spatial frequency included in the image 83 is ν [1 / mm], the focal length of the lens is f [mm],
If the wavelength is λ [mm], the corresponding Fourier transform image 8
4 appears at a position u from the center of the Fourier plane (position of 0th order light). Therefore, u needs to be in a range not exceeding the element. This is called a cutoff frequency and is expressed by the following equation.

U = λfν [mm] For example, ν = 40 [1 / mm], f = 400 [m
m] and λ = 500 × 10 ⁻⁶ [mm], u = 8
[Mm], and the two-dimensional optical element 21 has at least one side.
A length of 6 [mm] is required. On the contrary, ν = 40 [1 /
mm], λ = 500 × 10 ⁻⁶ [mm], u = 5 [mm]
Then, f = 250 [mm] and the focal length is 25
It is necessary to use a lens of 0 mm or less. Furthermore, f
= 400 [mm] and λ = 500 × 10 ⁻⁶ [mm], if u = [mm], only spatial frequencies up to ν = 25 [1 / mm] can be separated.

Therefore, the size of the two-dimensional optical element 21,
The focal length of the lens 60, the wavelength of the image 83, etc. may be appropriately determined.

Next, the type of mask pattern, fabrication, feature extraction and input will be described according to an embodiment of the present invention.

The mask pattern 30 is a real object illuminated by natural light or laser light, an image displayed on a display device such as a CRT display, and an image displayed on a spatial light modulator such as a liquid crystal panel and illuminated by light. Any image may be used as long as it has a light intensity distribution.

Any kind of image may be used as long as it has a light intensity distribution. Therefore, the easiest way to create a mask pattern is to place a printed matter, a photograph, an actual object, etc. at the position of the mask pattern 30 and to use white light. You can illuminate with. Further, a method of shooting the produced image with a CCD camera and displaying it on a CRT display, or a method of projecting it on a screen with a liquid crystal projector or the like may be used.

The mask pattern 30 is the image to be processed 10
It is a mask image which extracts the characteristic of. The extracted features change depending on the mask pattern shape. The relationship between the actual pattern shape and the feature will be described by a specific feature extraction example described later.

The mask pattern 30 is imaged and written as a real image input image 31 on the two-dimensional optical element 21 by the image forming lens 51 which is an image forming means. The lens 51 has a zoom function and a focusing function such as moving back and forth in the optical axis direction according to the size of the input image 30 and the distance, and the mask pattern 30 has an appropriate size on the writing surface side of the two-dimensional optical element 21. Image as an image of Especially,
When a three-dimensional real object is used as an input image, there is also an advantage that the focusing point by the focusing function makes clear the light and dark information of the image. The lens 51 also has a diaphragm function for adjusting the writing light amount. This diaphragm function is the same as the diaphragm function used for the image to be processed.

The control unit 100 according to the embodiment of the present invention is
The two-dimensional optical elements 20 and 21 are controlled, the erasing light of each two-dimensional optical element, and the reading light 81 are controlled.

Control when two two-dimensional optical elements 20 and 21 in FIG. 2 are combined will be described with reference to FIG. 7A is an electrical control pulse that the control unit 100 gives to drive the two-dimensional optical element 20 with E, W, and R as timings of erasing, writing, and reading, respectively, and FIG. (B) shows the control unit 10
Reference numeral 0 represents an electrical control pulse given to drive the two-dimensional optical element 21. On the other hand, FIG. 7C shows the control unit 1.
00 indicates the timing of controlling the light source and irradiating the reading light 81. The operation of the element at each timing is as described in the section for controlling the two-dimensional element.
The operation in the case where two two-dimensional optical elements are combined will be described using the 81 path of the reading light.

The light output from the light source 80 at the timing of FIG. 7C is linearly polarized by the polarization beam splitter 70, and the traveling direction thereof is bent by 90 degrees, and the two-dimensional optical element 20 is irradiated with the light. The light emitted to the two-dimensional optical element 20 is then reflected by the two-dimensional optical element 20, passes through the polarization beam splitter 70, and also passes through the polarization beam splitter 71, and is emitted to the two-dimensional optical element 21. Then, the light is reflected by the two-dimensional optical element 21 and is reflected by the polarization beam splitter 71.
It reaches the output surface 40 after being bent in the direction of 0 °. Although the above has been described step by step, since the light transmission time is extremely shorter than the operation time of the two-dimensional optical elements 20 and 21, it may be considered that there is no delay that should be considered in control.

Next, the output of the feature extraction image will be described according to an embodiment of the present invention. The lens 61 Fourier transforms the optical signal 86 corresponding to the logical product of the Fourier transform image 84 and the filter pattern 30, and forms the characteristics of the input image on the output surface as the output result 40.

Referring to FIG. 8, a series of operations for extracting the characteristics of the image to be processed as the output image according to the embodiment of the present invention will be described. This figure shows a specific example of an input image and an output image when two two-dimensional optical elements are combined. FIG. 8 (a)
Is written in the two-dimensional optical element 20, and the incoherent
A two-dimensional optical signal 83 is shown in FIG. 2, that is, a bright-dark image that has been coherently converted and carried into linearly polarized light of monochromatic light. The two-dimensional optical signal 83 is Fourier transformed by the lens 60 and becomes a Fourier transformed image 84 shown in FIG. 8B on the two-dimensional optical element 21. This becomes the reading light of the two-dimensional optical element 21. On the other hand, FIG. 8C shows the mask pattern 30 provided on the writing side of the two-dimensional optical element 21. In the crossed Nicols configuration using the polarization beam splitter 71, only the portion modulated by the writing light is output as a bright portion, so that the two-dimensional optical information 86 as shown in FIG. 8D is output. The two-dimensional light information 86 is Fourier-transformed by the lens 61, and the image shown in (e) of FIG. 8 is output to the output surface 40. This is the so-called "spatial filtering"
This is an operation corresponding to the contour extraction of the input image in.

Hereinafter, the operation of one embodiment of the present invention will be specifically described with reference to the drawings of the input image to be processed and the output image after feature extraction.

FIG. 9 is a diagram for concretely explaining the input image 10, the filter pattern 30, and the output result 40. 9A shows the processing target image 10. When the two-dimensional optical signal 84 converted into the coherent image is Fourier-transformed, a Fourier-transformed image as shown in FIG. 9B is obtained on the two-dimensional optical element 21. can get. However, FIG. 9B is a schematic representation, and since the image size and the like differ depending on the focal length of the Fourier transform lens and the wavelength of light, the accurate conversion result with respect to FIG. 9A is obtained. Does not indicate. Here, if a two-dimensional optical element 21 in FIG. 2 is written with an image that allows the Fourier transform image to pass over the entire surface as the mask image 30 (all bright), the Fourier transform image shown in FIG. 9B is obtained. The image is represented by (c) in FIG. 9 on the output surface 40 after being Fourier-transformed again by the lens 61. This is an inverted image.

The image shown in FIG. 9 will be described in more detail. However, in FIGS. 9 to 12, the spatial frequencies included in the processing target image from which the features are extracted are almost equal (within several times). Typical examples of such images include images in which contour information is extracted in advance from characters, line images, or natural images. A case of handling an image including various spatial frequencies will be described with reference to FIG.

When the image shown in FIG. 9A is input as the processing target image 10 to the image processing apparatus of one embodiment of the present invention, while the image shown in FIG. 9D is input as the mask pattern 30. , The image shown in (e) of FIG. 9 is output.
This is a result of the schlieren effect strongly appearing because the 0th-order light of the Fourier transform image is blocked and the contour extraction is performed. In addition, in order to simplify the description, upside down, leftward and rightward inversion is omitted.

Next, when the image shown in FIG. 9A is input as the input image 10 to the optical image processing apparatus of the present invention, while the image shown in FIG. 9F is input as the mask pattern 30, The image shown in FIG. 9G is output. This is a result of smoothing the image by partially losing the input shape information in order to block higher-order components of the Fourier transform transformed image.

FIG. 10A shows the image 10 to be processed. When the two-dimensional optical signal 83 converted into a coherent image is Fourier-transformed, the Fourier transform as shown in FIG. The image appears. However,
(B) of 0 is a schematic representation. FIG. 1 shows a processing target image 10 in the image processing apparatus according to the embodiment of the present invention.
When the image shown in (a) of 0 is input and the image shown in (c) of FIG. 10 is input as the mask pattern 30, the image shown in (d) of FIG. 10 is output. This is the result of extracting a line segment in the vertical direction (horizontal direction) with respect to this specific direction in order to block a portion other than the specific direction (vertical direction) of the Fourier transform image. In order to simplify the description, the up / down / left / right inversion of the output image is omitted.

Similarly, the processing target image 1 is shown in FIG.
When 0 is input and the image shown in (e) of FIG. 10 is input as the mask pattern 30, the image shown in (f) of FIG. 10 is output. Further, when the image shown in (g) of FIG. 10 is input as the mask pattern 30, the image shown in (h) of FIG. 10 is output. When the image shown in (i) of FIG. 10 is input as the mask pattern 30, the image shown in (j) of FIG. 10 is output.

As can be seen from the above results, the direction of the line segment output changes with the direction of the mask pattern. Therefore, when the mask is continuously rotated about the center of the mask, directional line segments are obtained in time series. By storing this series of processing in another recording element at a certain timing,
For example, a combination of vertical and horizontal directional line segments can be obtained. As an example of a method of rotating the mask, a method of rotating an actual object or controlling by an electric projection means is possible.

The circle in FIG. 11A or FIG. 11B is the image to be processed 10. When the two-dimensional optical signal 83 converted into a coherent image is Fourier-transformed, both are two-dimensional optical element 21. A concentric Fourier transform image as shown in FIG. 11C appears above. However, FIG. 11 (c)
Is a schematic representation.

The image shown in FIG. 11A or FIG. 11B is input as the processing target image 10 to the image processing apparatus according to one embodiment of the present invention, and the mask pattern 30 is input to the image shown in FIG. 11D. When the image shown is input, the image shown in (e) of FIG. 11 is output. This is the result of extracting only the strictly specific direction (horizontal direction) line segment as shown in (f) of FIG. 11 in order to block the parts other than the specific direction (vertical direction) of the Fourier transform image. .

On the other hand, when the image shown in (g) of FIG. 11 is input as the mask pattern 30, the image shown in (h) of FIG. 11 is output. This is selected by giving an appropriate range in the direction of the Fourier transform image, and in order to block other parts, a line segment group having a range corresponding to the filter image as shown in (i) of FIG. 11 is output. It means that.

FIG. 12 is a diagram showing extraction of directional line segments from an image that is not a line drawing. 12A shows the processing target image 10. When the two-dimensional optical signal 83 converted into the coherent image is Fourier-transformed, a Fourier-transformed image as shown in FIG. 12B appears on the two-dimensional optical element 21. . However, FIG. 12B is a schematic representation.

When the image shown in FIG. 12A is input as the processing target image 10 to the image processing apparatus according to the embodiment of the present invention, while the image shown in FIG. 12C is input as the mask pattern 30. , The image shown in (g) of FIG. 12 is output. This is a result of extracting a line segment in a direction (horizontal direction) perpendicular to the specific direction in order to block 0-order light from a portion other than the specific direction (vertical direction) of the Fourier transform image.
Similarly, when the image shown in (d) of FIG. 12 is input as the mask pattern 30, the image shown in (h) of FIG. 12 is output.

As described above, in order to obtain the directional line segment, the 0th-order light of the Fourier transform image is blocked and the part other than the Fourier transform image corresponding to the necessary directional line segment is blocked. However, the figure shown in (e) of FIG. 12 or (f) of FIG. 12 may be used as a filter for performing such blocking.

A so-called "spatial filtering" will be described below as a specific feature extracting process of the image processing apparatus according to the embodiment of the present invention.

The principle of spatial filtering will be described with reference to FIG. This figure is an explanatory diagram in which a basic re-diffraction system is configured by using two convex lenses. The focal length of the first Fourier transform lens is f1, the focal length of the second Fourier transform lens is f2, and the two lenses are arranged as shown in FIG. When a two-dimensional amplitude modulation image 2 is placed on the input surface 1 and coherent light 3 is irradiated from behind, the image 2 is diffracted by the Fourier transform lens 4, and the Fourier transform image of the image 2 appears on the Fourier plane 5 at the distance f1. Further, it is re-diffracted by the second Fourier transform lens 6 installed at a position at a distance f2 from the Fourier plane 5, and an inverted image 8 of the input image appears on the output plane 7. If an amplitude modulation image 9 (this is referred to as a spatial filter) that allows a part of the Fourier transform image to pass through and cuts off a part of the Fourier transform image on the Fourier plane 5 is selected, the optical signal can be selected according to the shape of the filter and the transmittance. Done and printed.

The above spatial filtering is used to improve the performance of images (particularly photographs), and is specifically applied as follows.

Noise removal, which is a typical example of optical image processing by filtering, improves the image by arranging a spatial filter corresponding to the Fourier transform of the noise when the image contains periodic noise. This is used for removing moire fringes in printed matter.

The high-pass filtering cuts off a part of the low band (0th order light) of the Fourier transform image of the image when the image has low contrast or when the image is unclear due to slight defocus or aberration. By installing such a spatial filter, it is possible to increase the gain in the high frequency region and make the “edge” of the image more noticeable. 0 in extreme cases
If a spatial filter that blocks all secondary light and passes only higher-order components is installed, contour extraction of the image is executed.

The low-pass filtering is performed when the image contrast is too strong, or when the image contains fine noise, the Fourier transform image of the image is in the low range (from 0th order light to 3rd order light or 5th order light). A spatial filter can be installed to allow smoothing or denoising of the input image.

Bandpass filtering is processing by a spatial filter that allows a specific frequency region of the spatial frequencies included in the image to pass. A filter that allows an annular portion to pass through the central portion (0th order light) of the Fourier transform image is used.

The optical image processing using the Fourier transform-inverse transform described above is described in detail in, for example, "Optical Technique Handbook, Supplementary Edition", published by Asakura Shoten.

Hereinafter, "spatial filtering" for handling an image including various spatial frequencies by the image processing apparatus according to the embodiment of the present invention will be described with reference to FIG. The application is to improve image quality. FIG. 13 is a diagram for explaining the image to be processed, the filter pattern, and the result.

A typical example of optical image processing by filtering is noise removal for improving an image by installing a spatial filter corresponding to the Fourier transform of the noise when the image contains periodic noise. .

FIG. 13A shows an input image in which the light intensity of the image to be processed at the point (x, y) is represented by g (x, y) with the x axis in the horizontal direction and the y axis in the vertical direction. Is. That is, in FIG. 13A, a white portion indicates a portion with light (strong portion) and a black portion indicates a portion without light (weak portion, blocked portion). As shown in (b) of FIG. 13, it is assumed that noise that does not transmit light at all appears in a cycle of 0.1 mm in the x-axis direction, and the intensity of light during that time changes sinusoidally. The intensity I (x, y) of this noise light at the coordinates (x, y) is I (x, y) = 1 + cos
It is represented by (20πx). When an image in which this noise is superimposed on the image to be processed is used as an input image of the present apparatus, Fourier transform images of g (x, y) and I (x, y) are superimposed and appear on the Fourier transform plane. The focal length of the Fourier transform lens 60 is f = 400 [mm], and the wavelength of the two-dimensional optical signal 83 is λ =
It is set to 500 × 10 ⁻⁶ [mm]. The pitch of the dark line is 0.1
Since it is mm, the spatial frequency ν = 10 [lines / mm]. From the above numerical values, the noise component is concentrated in the part where u = λfν = 2 [mm]. Therefore, the mask pattern 30 is shown in FIG.
3 (c), the imaging magnification of the imaging lens 51 is adjusted so that the light-impermeable portion is located at a distance of 2 mm from the center on the two-dimensional optical element 21, and writing is performed on the optical element. When this is done, spatial filtering is performed to obtain the image from which the noise shown in FIG. 13 (e) has been removed.
Note that the mask pattern 30 shown in FIG. 13D may be used instead of FIG. 13C.

The high-pass filtering by the image processing apparatus according to the embodiment of the present invention is performed by the mask pattern 30 which blocks a part of the low band (0th order light) of the Fourier transform image of the image shown in FIG. It is realized by using.

The low-pass filtering by the image processing apparatus according to the embodiment of the present invention is performed by the low band of the Fourier transform image of the image shown in (g) of FIG. 13 (appropriately from 0th order light to 3rd order light or 5th order light). Pattern 30 for passing through
It is realized by using.

The bandpass filtering by the image processing apparatus according to the embodiment of the present invention is realized by using the mask pattern 30 for transmitting the annular portion of the Fourier transform image shown in (h) of FIG. It On the contrary, the process of blocking the specific frequency region is realized by using the mask pattern 30 that blocks the annular portion of the Fourier transform image shown in (i) of FIG.

[0113]

As described above, since the image processing apparatus for feature extraction of the present invention can process an object image illuminated by natural light by an optical system, the image processing is performed on a moving object in real time. It can be performed. Further, the image used as the mask can be easily produced, for example, by using the screen of the CRT, and can be changed at any time in time series. Further, it can be constituted by an optical device such as liquid crystal that consumes little energy and an optical component that causes little loss of optical signals, and an extremely simple and high-speed optical image processing device can be realized. Further, since the feature can be extracted in real time while checking the result by the means for adjusting the light intensity of the input image, a function such as noise removal can also be realized. Furthermore, the contour extraction device of the present invention can be connected in parallel to extract the feature relating to the color information.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a system configuration diagram of an embodiment of the present invention.

FIG. 3 is a diagram illustrating a two-dimensional optical element, in which FIG.
The structure of a three-dimensional optical element is shown, (b) shows a rubbing direction.

FIG. 4 is a diagram showing control pulses of a two-dimensional optical element according to an embodiment of the present invention.

FIG. 5 is a diagram showing an operating principle of a two-dimensional optical element.

6A and 6B are diagrams showing an input / output image of the two-dimensional optical element of one embodiment of the present invention, in which FIG. 6A is a writing image, FIG. 6B is a reading light, and FIG.

FIG. 7 is a diagram showing control pulses of two two-dimensional optical elements according to an embodiment of the present invention.

FIG. 8 is a diagram showing a series of images in the feature extraction processing according to the embodiment of the present invention.

FIG. 9 is a diagram showing a processed image of contour extraction and smoothing according to an embodiment of the present invention.

FIG. 10 is a diagram showing a processed image of directional line segment extraction according to an embodiment of the present invention.

FIG. 11 is a diagram showing a processed image of a directional line segment extraction having a range of directionality according to an embodiment of the present invention.

FIG. 12 shows a processed image for directional component extraction of an image having a width according to an embodiment of the present invention.

FIG. 13 is a diagram showing a noise removal processed image according to an embodiment of the present invention.

FIG. 14 is an explanatory diagram showing the principle of spatial filtering.

[Explanation of symbols]

 1 Input Surface 2 Input Image 3 Coherent Light 4,6 Fourier Transform Lens 5 Fourier Transform Surface 7 Output Surface 8 Filtering Result 9 Spatial Filter 10 Image to be Processed 20,21 Two-dimensional Optical Element 30 Mask Pattern 31 Mask Image Real Image 40 Feature Extraction Result 50,51 Imaging lens 60,61 Fourier transform lens 70,71 Polarization beam splitter 80 Light source 81 Readout light 82 Modulated two-dimensional optical signal 83 Coherent light image 84 Fourier transform image of coherent light image 85 Two-dimensional light of image feature Signal 86 two-dimensional optical signal which is a logical product 100 control unit 200 image input means for processing 210 first polarization modulation means 220 Fourier transform means 230 mask input means 240 second polarization modulation means 250 control means 260 processed image output means 270 light source

Claims (2)

[Claims] 1. A feature extraction image processing apparatus for masking a part of an image obtained by performing an optical Fourier transform on a processing target image, and performing optical Fourier transform on the masked image again to perform feature extraction of the processing target image, Polarization modulation means for receiving the reading light and the writing light and changing the polarization angle of the reading light according to the light intensity of the writing light, and the writing of the light of the mask image used for the mask processing to the polarization modulating means. Means for supplying the light as the light, means for optical Fourier transforming the light of the image to be processed into the readout light to the polarization modulating means, and output light of the polarization modulating means, which is polarized at a predetermined angle. An image processing device for feature extraction, comprising: processed image output means for transmitting light, taking out masked light, and performing optical Fourier transform of the light to output. 2. A feature extraction image processing apparatus for masking a part of an image obtained by performing an optical Fourier transform on a processing target image, and performing optical Fourier transform on the masked image again to extract a feature of the processing target image, First polarization modulation means for receiving the read light and the write light and changing the polarization angle of the read light according to the light intensity of the write light; and the light intensity of the write light for receiving the read light and the write light. Second polarization modulation means for changing the polarization angle of the readout light in accordance with the first polarization modulation means, means for using the light of the image to be processed as the writing light to the first polarization modulation means, and the first polarization modulation means. Means for optical Fourier transforming the output light of the means to obtain the readout light to the second polarization modulation means, and the writing of the light of the mask image used in the mask processing to the second polarization modulation means. A means for supplying light as a light and a processing for passing the light polarized at a predetermined angle out of the output light of the second polarization modulation means to take out the masked light and subject it to an optical Fourier transform for output. An image processing apparatus for feature extraction comprising image output means and control means for controlling the operations of the first and second polarization modulation means.