JPH0843768A - Image processor - Google Patents

Abstract

PURPOSE:To obtain an image processor capable of accurately discriminating and recognizing an image which is shifted, rotated and deformed, such as enlarged or reduced, at high speed. CONSTITUTION:The image information of an inputted substance worked to invariant information by Fourier-transforming and coordinate transforming the image by means of a Fourier transformation optical system 1 and a coordinate transformation optical system 2 and shifting, rotating and deforming, such as enlarging or reducing, the image is instantaneously fetched as plural pieces of spatial frequency information the most effective to discriminate its characteristic by a discrimination optical system 3, and further recognized by a neural network being a recognition system 4, so that the image which is shifted, rotated and deformed, such as enlarged or reduced, is accurately discriminated and recognized at high speed.

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,
In particular, the present invention relates to an image processing device capable of identifying and recognizing images that have undergone deformation such as shift, rotation, and enlargement / reduction.

[0002]

2. Description of the Related Art There is a strong need for a device for recognizing an object using an image, such as a visual device for an industrial robot or a product inspection device for an automation line, and a correlator using an optical matched filter has been conventionally used. The development of a recognition device has been actively carried out. Further, recently, development of a recognition device using a neural network has been actively conducted. However, each of these recognition devices has a low ability to recognize an image that has undergone deformation such as shift, rotation, and enlargement / reduction, that is, the generalization ability is low, and improvement of this generalization ability is not practical. Had been the biggest challenge of.

One answer to this problem was Carn.
David Casasent of egie Mellon Univ. etc. (David
Casasent et.al. "Real-time deformation invariant o
ptical pattern recognition using coordinate tranfo
rmations ", Appl. Opt., Vol.26, pp.938-942 (1987)). The method of using their correlators is as follows. First, as shown in FIG. The image 101 is illuminated with coherent light 102, and a computer generated hologram (CGH) 10 is applied to the image 101.
When the Fourier transform lens L ₁ 104 performs the Fourier transform after superimposing the phase information for the coordinate transform by 3, the information obtained by performing the desired coordinate transform on the input image is obtained on the coordinate transform surface 105. In the case of the above-mentioned document, for example, the desired coordinate transformation is a logarithmic polar coordinate transformation in which even if transformations of scaling and rotation occur in the input image, those transformations are transformed into shift amounts.

Next, as shown in FIG. 8, the information after the logarithmic polar coordinate conversion is applied to a liquid crystal television (LCTV: liquid crysta).
a double diffractive system consisting of two Fourier transform lenses L ₂ 107 and L ₃ 108 and a matched filter (MSF: MSF: Matched spatial filter) 1
The coherent light 110 is incident on the shift invariant correlation optical system in which the filtering by the 09 is combined, and the correlation surface P is obtained by the camera 111.
It is to detect the correlation peaks of the reference image and the image to be detected and recognize them. In fact, Casasent et al. In the above-referenced document show an example in which good matching results are obtained for rotation and scaling deformations using the above method.

Further, the US Air Force Institute of Tech
Although there is no detailed explanation, Mr. Kenneth H Fielding of nology develops this and transforms the input image into a Fourier spectrum (square of the Fourier transform of the input image), and then performs the same processing to rotate and In addition to enlarging / reducing, it has been shown that good matching results were obtained regarding shift deformation (Kenneth H. Fielding et.a.
l. "Position, scale, and rotation invariant hologr
aphic associative memory ", Opt. Eng., Vol.28, pp.849
-853 (1989)).

[0006]

The method of Mr. Fielding et al., Which tried to develop the method of Mr. Casasent et al. To have generalization ability with respect to shift deformation, is the same as that of Fourier spectrum of the input image. However, the similarity between the images has already increased at the time when the input image is transformed into the Fourier spectrum, and the image after the coordinate conversion is the same as the input image. Compared to the ones, there are many that are quite similar. Therefore, it is actually difficult to identify or recognize the image having a high degree of similarity with a simple Vander-Lugt type correlation optical system like Mr. Fielding.
Furthermore, this problem is solved by improving the accuracy of the correlation optical system (example: JTC
It may be possible to solve the problem by changing to, or using the phase type, etc.), but this does not show sufficient improvement.

The present invention has been made in view of such circumstances, and an object thereof is image processing capable of accurately and rapidly identifying and recognizing an image which has been deformed such as shift, rotation, and enlargement / reduction. It is to specifically provide the device.

[0008]

The image processing apparatus of the present invention for achieving the above object comprises a first image inputting means for inputting an image to be recognized into the system, and the first image inputting means. The first Fourier transform lens for converting the image information input into the system by the means into the Fourier spectrum information, and the first Fourier transform lens are output to the Fourier transform plane placed at the position of the rear focal point of the first Fourier transform lens. Fourier transform optical system including first detecting means for detecting Fourier spectrum information, second image inputting means for inputting the detected Fourier spectrum information into the system, and second image inputting means. A computer generated hologram for superimposing phase information for desired coordinate conversion on the Fourier spectrum information of the image input into the system by the image input means of, and further, this superimposed information. Second Fourier transform lens for Fourier transform, and second detecting means for detecting coordinate transform information output to the coordinate transform surface placed at the position of the rear focal point of the second Fourier transform lens. A coordinate conversion optical system, a third image input means for inputting the detected coordinate conversion information into the system, and coordinate conversion information input into the system by the third image input means. A multiplexing means for multiplexing, the filtering means for filtering the multiplexed coordinate transformation information in different spatial frequencies in the individual areas, and the coordinates in the individual areas spatially frequency filtered by the filtering means. An identification optical system including third detection means for detecting conversion information, and a recognition system for recognizing an input image based on an output detected by the identification optical system. It is characterized in that comprises.

In this case, a two-dimensional lens array is used as the multiplexing means, and a system including a two-dimensional Fourier transform lens array and a two-dimensional spatial frequency filter array arranged on the Fourier transform plane is used as the filtering means. be able to.

As the multiplexing means, a system provided with an image-forming optical system and a two-dimensional diffraction means arranged in the system is used, and as a filtering means, a two-dimensional Fourier transform lens array and its Fourier transform surface are used. A system with a two-dimensional spatial frequency filter array arranged can also be used.

Further, the multiplexing means has an optical system for irradiating the collimated light from a plurality of different directions with respect to the coordinate conversion information inputted into the system by the third image input means, and the Fourier transform means as the filtering means. It is also possible to use a system with a transform lens and a two-dimensional spatial frequency filter array arranged on its Fourier transform plane. In this case, the spatial light modulator of the optical writing type and the reflection type may serve as the second detecting means and the third image inputting means.

[0012]

In the present invention, the spatial frequency information of the information which is the most effective for clearly distinguishing the features of the Fourier-transformed and coordinate-transformed information is simultaneously extracted by a discriminating optical system in a plurality of channels, and the information is extracted. Since the recognition system is used to recognize, it is possible to accurately and quickly identify and recognize even an image that has been deformed by shifting, rotating, scaling, or the like.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The image processing apparatus of the present invention will be described below in more detail based on the preferred embodiments with reference to the drawings. [Embodiment 1] An image processing apparatus of this embodiment will be described with reference to the diagram showing the configuration of FIG. In this apparatus, the Fourier transform optical system 1 includes a first image input unit including an image input unit 11 and a first illumination unit 12 for inputting an image including an object to be recognized into the system, and the first image input unit. And a first Fourier transform lens 13 for performing Fourier transform on the object information, and a first Fourier transform lens 13 for detecting the Fourier spectrum information output to the Fourier transform surface at the rear focal position of the first Fourier transform lens 13. It is composed of a screen 14 placed on this Fourier transform plane as a detecting means and a two-dimensional image pickup device 15, and transforms the object information to be recognized into shift invariant Fourier spectrum information, and the coordinate transformation optics of the next stage. It allows the system to communicate this information.

First, the image input device 11 includes an image pickup device 11a for inputting an image including an object to be recognized into the system.
And a spatial light modulator 11b for displaying the object information imaged by the imaging element 11a in the system, and a driver 11c for driving the spatial light modulator 11b. Specifically, the image pickup device 11a may be a tube type or a solid-state image pickup device, as long as it can pick up an image, but in this embodiment, a CCD camera with a zoom lens is used. I was there. Further, an image picked up by the CCD camera which is the image pickup device 11a is converted into a TV signal such as NTSC by a circuit in the camera, but in the present embodiment, the spatial light modulator 11b is used as a CRT of a TV or a computer. Using a transmissive liquid crystal element that has recently come to be used, the driver 11c takes in this signal and displays it in the Fourier transform optical system 1. The displayed object information is the first illumination means 12
Is illuminated by a substantially parallel coherent light beam and is input into the system. Of course, at this time, if the diffraction image of the liquid crystal element 11b interferes, it can be easily removed by a spatial frequency filter or the like. The coherent light beam which is the first illuminating means 12 makes the light from the coherent light source 12a into a substantially parallel light beam by the beam expander 12b. As the coherent light source 12a, various lasers,
Although various lamps such as a halogen lamp with a filter can be used, in the present embodiment, an Ar laser of 514.5 nm was used. Also, the beam expander 12b
Is a conventional means, but it is constructed by using a diaphragm lens 12b ₁ , a pinhole 12b ₂ and a collimator lens 12b ₃ . This coherent light beam is reflected by the mirror 12c and the beam splitter 1 in order to make the entire apparatus compact.
The light flux is bent using 2d and is made incident on the system, but these are not essential in the configuration. In addition, in the first detection means, the screen 14 placed on the Fourier transform plane is a diffuser having a diffusion surface, and the two-dimensional image pickup device 15 is a tube type or a solid-state image pickup device. Anything can be used as long as it can be used, but a CCD camera similar to that of the image pickup device 11a is used.

Next, the coordinate transformation optical system 2 is a second image composed of an image input device 21 and a second illumination means 22 for inputting the Fourier spectrum information transformed by the Fourier transformation optical system 1 into the system. Input means and CGH (computer hologram) 23 for superimposing phase information for logarithmic polar coordinate conversion (Logr-θ conversion) on Fourier spectrum information of an image input into the system by the image input means.
Further, the second Fourier transform lens 24 for Fourier transforming the superimposed information, and the coordinate transform information output to the coordinate transform surface at the rear focal position of the second Fourier transform lens are detected. It is composed of a screen 25 placed on this coordinate conversion surface, which is a second detection means, and a two-dimensional image pickup device 26, and converts the object information to be recognized into invariant information for rotation and scaling. However, this information is transmitted to the identification optical system in the next stage.

Here, the image input device 21 is a spatial light modulator 21 for displaying information sent from the two-dimensional image pickup device 15 of the first detecting means of the Fourier transform optical system 1 in the system.
a and a driver 21b for driving the spatial light modulator 21a. Specifically, the spatial light modulator 21
As a, a transmissive liquid crystal element similar to the spatial light modulator 11b was used, and an image was displayed in the coordinate conversion optical system 2.
The displayed object information is illuminated by the substantially parallel coherent light flux which is the second illumination means 22, and is input into the system. In the case of this embodiment, this coherent light beam 22 is the light beam 12 used in the Fourier transform optical system 1 for simplicity.
To use. Specifically, the light beam transmitted through the beam splitter 12d is bent by the beam splitter 12e and used as a substantially parallel coherent light beam. In addition, as the second detecting means, the screen 25 placed on the coordinate conversion surface is a diffuser having a diffusion surface similar to that of the screen 14, and the two-dimensional image pickup device 26 is a tube type solid-state image pickup device. However, as long as it can pick up an image, any CCD camera similar to the two-dimensional image pickup device 15 is used.

Actually, when the apparatus configured as described above was used to create an image in which various images centered on the alphabet of characters were Fourier-transformed and the coordinates were further transformed, even if the original images differ greatly, Many of the images were relatively similar images. Further, in order to distinguish these similar images, the optimum features were examined. As a result, many of these images were composed of line segments or rectangular elements along the Logr axis and the θ axis. It has been found that it is good to distinguish them by spatial frequency features along one axis. Therefore, the identification optical system 3 includes an image input device 31 and a third illumination means 32, which are third image input means for inputting the information converted by the coordinate conversion optical system 2 into the system, and the third optical input means 31. 3 means for multiplexing the information input into the system by the image input means, 3) filtering means 34 for filtering the multiplexed information in different spatial frequencies in the respective areas, and 3rd detection The means 35 is used to instantly detect the characteristics at a plurality of spatial frequencies of the information that has been Fourier transformed, further coordinate transformed, and processed into invariant information for shift, rotation, and deformation of scaling, and recognizes the next stage. It allows the system to communicate this information.

Here, the image input device 31 is a spatial light modulator 31a for displaying information transmitted from the two-dimensional image pickup device 26 of the second detecting means of the coordinate conversion optical system 2 in the system.
With the driver 31b for driving the spatial light modulator 31a, the third illuminating means 32 is constituted by a backlight illuminating for generating an incoherent light beam. Specifically, as the spatial light modulator 31a, the spatial light modulator 21a is used.
An image is displayed in the system of the identification optical system by using a transmissive liquid crystal element similar to the above, and is further illuminated by the backlight illumination 32 and input into the system. Next, the multiplexing means 33 is composed of a collimator lens 33a and a two-dimensional lens array 33b, and creates a plurality of duplicate images of the input image at the back focal position of the two-dimensional lens array 33b. At this time, the collimator lens 33a is installed so that the display surface of the spatial light modulator 31a is located at the front focus position thereof, and the optical axis of each lens of the two-dimensional lens array 33b is parallel to the optical axis of the collimator lens 33a. To install. further,
The filtering means 34 is a two-dimensional lens array 33b.
In the filtering means, the writing surface is installed so as to coincide with the rear focal point position, a plurality of pieces of information created by the multiplexing means are written, and further read out by the coherent light beam which is the fourth illuminating means 34a. The optical writing type reflective spatial light modulator 34b for inputting the read information as coherent information, the driver 34c for operating the spatial light modulator 34b, and the plurality of read coordinate transformations. A two-dimensional Fourier transform lens array 34d installed so that its front focus position for Fourier transforming the image for each individual region thereof coincides with the reading surface of the spatial light modulator 34b, and for each individual region thereof. In order to perform different spatial frequency filtering, a two-dimensional spatial frequency filter installed at the back focal position of this two-dimensional Fourier transform lens array 34d. It is composed of a Rutaare 34e.

The spatial light modulator 34b includes a crystal spatial modulator such as bismuth silica oxide or barium titanate, an organic nonlinear material spatial modulator, and a refraction / deflection type spatial modulator such as a deformable mirror. Although various devices such as a liquid crystal device can be considered, a liquid crystal spatial modulator is used in this embodiment. In the case of a liquid crystal spatial modulator, if it is a ferroelectric type or a super twisted nematic type, the recording / reproducing is 10
With a high-speed cycle of 0 to 2000 Hz,
It is possible with a high resolution of 100 to 200 lines / mm.
In addition, the two-dimensional spatial frequency filter array 34e has the same number of filters as the two-dimensional Fourier transform lens array and the filters of the same array are arranged, and as shown in FIG.
About half of them correspond to the Logr axis, and the rest correspond to the θ axis. Each filter has an opening provided at a different distance from the optical axis position of the corresponding lens of the two-dimensional Fourier transform lens array 34d. Further, the third detecting means 35 is composed of a two-dimensional detector array installed adjacent to the two-dimensional spatial frequency filter array 34e, and the light receiving portion of each detector is a two-dimensional spatial frequency filter. Array 34e
It has the same size that can cover the opening position of each filter. Further, as the coherent light beam which is the fourth illuminating means 34a, in the case of the present embodiment, the light beam 12 used in the Fourier transform optical system 1 is used for simplicity. Specifically, the light beam that has passed through the beam splitter 12e is bent by the mirrors 12f and 12g and the beam splitter 12h, and is used as a substantially parallel coherent light beam 34a.

Finally, the recognition system 4 is Fourier-transformed by the Fourier-transform optical system 1 and further coordinate-transformed by the coordinate-transforming optical system 2 so as to be processed into invariant information for shift, rotation and scaling deformation. The information thus obtained is further inputted as a feature at a plurality of spatial frequencies that are instantaneously detected by the identification optical system 3 for recognition. In this embodiment, the recognition system 4 uses a back propagation model of a neural network. The back propagation model of the neural network of this embodiment is realized as software on a computer.

In this back propagation model,
First, it is necessary to determine the synaptic weight between each layer of the neural network by learning. This is processed and detected by sequentially inputting an image to be recognized into the apparatus by the first image inputting means, and passing through the Fourier transform optical system 1, the coordinate transform optical system 2 and the identification optical system 3 described above. , The third detection means of the discrimination optical system 3 instantaneously detects features at a plurality of spatial frequencies, and inputs these features to the neurons of the input layer of the neural network. At this time, the features of the output layer of the neural network are input. The backpropagation learning rule may be used to determine the synaptic weights between layers so that the firing of neurons is selectively performed corresponding to the image to be recognized. At the time of recognition, the synapse load between the layers may be fixed, the image to be recognized may be input into the apparatus by the first image input means, and firing of neurons in the output layer may be detected. With the above configuration, the image processing apparatus
Even if the input image is shifted, rotated, or deformed by scaling,
Obviously, they can be recognized as the same object.

In the above embodiment, a backpropagation model neural network was considered as the recognition system 4, but the same applies when other neural network models such as Hopfield model, RCE model and ART model are used. It goes without saying that an image processing apparatus having various effects can be realized. Further, in the above embodiment, the neural network used as the recognition system 4 was constructed by software on a computer, but it is of course possible to use a dedicated neurochip or an optical device. Further, as the recognition system 4, a correlator using a matched special filter, a joint transform correlator (JTC), or the like can be used to realize an image processing device having a similar effect without using a neural network.

Further, in the above embodiment, the two-dimensional spatial frequency filter array 34e is an element in which an opening is further provided at a position symmetrical with respect to the center (optical axis) as shown in FIG. 2 (b). It may be configured, or, as shown in FIG. 2C, may be configured by an element having an asymmetric opening. Further, as shown in FIG. 2A, it is not necessary that the openings are arranged at uniform intervals, and it is of course possible to arrange the areas corresponding to the particularly important frequency components fine and the other areas coarsely arranged.

Further, in the above embodiment, the multiplexing means 3
3 is not limited to the combination of the single collimator lens 33a and the two-dimensional lens array 33b shown in FIG.
As shown in (a), a combination of a single collimator lens 33a and a refractive index distribution type microlens array 33b ', and as shown in (b) of the same figure, a single collimator lens 33a and a refractive index distribution type The combination of the rod lens array 33b ″, the combination of the gradient index rod lens 33a ′ as the collimator lens and the gradient index microlens array 33b ′ as shown in FIG. Various types can be used.

[Embodiment 2] The image processing apparatus of the present embodiment is
This will be described with reference to the diagram showing the configuration of FIG. In this embodiment, the third illuminating means 32 and the multiplexing means 33 of the third image input means in the identification optical system 3 of the first embodiment have different configurations, and the other configurations are the same as those of the first embodiment. In the figure, for simplification, only the identification optical system 3, the recognition system 4 and the parts associated therewith are shown.

In the identification optical system 3 of this embodiment, the image of the image input device 31 for displaying the information processed through the Fourier transform optical system 1 and the coordinate transform optical system 2 in the system is the third illumination. The substantially parallel coherent light flux, which is the means 32, is input into the system. In this embodiment, the light flux 12 used in the Fourier transform optical system 1 is used for simplicity. Specifically, the light beam transmitted through the beam splitter 12e is bent by the beam splitter 12m and is used as a substantially parallel coherent light beam which is the illuminating means 32. The multiplexing means 33 is arranged so that the lens 33c and the lens 33d are confocal, and the display surface of the spatial light modulator 31a is located at the front focus position of the lens 33c on the front side of the lens 33c. In the case of the present embodiment, a Dammann type two-dimensional grating is installed as the two-dimensional grating 33e at the focal position (because of the confocal arrangement, it coincides with the position of the front focal point of the lens 33d). With this configuration (so-called double-diffraction system arrangement), a multiple-focused image is formed at the rear focal point of the lens 33d. The multiple-imaged information is further sent to the recognizing means 4 via the filtering means 34 and the third detecting means 35. However, as in the first embodiment, the input image is deformed by shifting, rotating or scaling. It goes without saying that an image processing apparatus that can recognize the same object even if received can be configured.

[Third Embodiment] An image processing apparatus according to the present embodiment is
This will be described with reference to the diagram showing the configuration of FIG. In this embodiment, the identification optical system 3 has a simpler structure,
The other parts are configured the same as in the first embodiment. For the sake of simplicity, only the identification optical system 3, the recognition system 4 and the parts associated therewith are shown in the drawing.

The identification optical system 3 of this embodiment is an image input device 31 which is a third image input means for displaying the information processed through the Fourier transform optical system 1 and the coordinate transform optical system 2 in the system. And a substantially parallel coherent light flux and multiplexing means 33, which is a third illumination means 32 for generating collimated light from various directions to read out the displayed image, and superposition of the read information. Each of the regions formed by the collimated light in each direction is composed of filtering means 34 for performing different spatial frequency filtering and third detecting means 35, and Fourier transform and coordinate transformation are performed to shift, rotate, and enlarge / reduce. The characteristics of information processed into invariant information at multiple spatial frequencies are instantly detected, and this information is transmitted to the recognition system 4 in the next stage. It is intended.

Here, as the substantially parallel coherent light beam which is the illuminating means 32, the light beam 12 used in the Fourier transform optical system 1 is used for simplicity in this embodiment. Specifically, the light flux obtained by bending the light flux transmitted through the beam splitter 12e by the mirror 12i is used as the illuminating means 32 and is a substantially parallel coherent light flux. This lighting means 3
The substantially parallel coherent light flux of 2 is the multiplexing means 33.
The collimated light from various directions is converted by the refractive index distribution type microlens array 33b ′ and the collimator lens 33a which are arranged in the confocal arrangement. The two-dimensional lens array 33b 'is installed so that the optical axis of each lens is parallel to the optical axis of the collimator lens 33a. Information sent from the two-dimensional image pickup device 26 of the second detecting means of the coordinate conversion optical system 2 is superimposed on the collimated light from various directions by the image input device 31. Reference numeral 31 includes a spatial light modulator 31a and a driver 31b for driving the spatial light modulator 31a. Specifically, as the spatial light modulator 31a, a transmissive liquid crystal element similar to that of the first embodiment is used, and its display surface is arranged so as to coincide with the rear focal position of the collimator lens 33a. Further, the collimated light in each direction in which the read information is superimposed is the Fourier transform lens 34d ′.
And a two-dimensional spatial frequency filter array 34e similar to that of the first embodiment. Specifically, the collimated light in each direction in which the read information is superimposed is Fourier-transformed by the Fourier transform lens 34d ′ installed so that the front focus position matches the display surface of the spatial light modulator 31a. A plurality of pieces of Fourier transform information (of superposed information) corresponding to the collimated light in each direction are formed at the rear focal position of the Fourier transform lens 34d '. For simplicity, only the behavior of the light flux with respect to this one element is shown by hatching in the figure. Then, this Fourier transform lens 34d '
If a two-dimensional spatial frequency Fourier array 34e similar to that of the first embodiment is installed at the rear focal position, different spatial frequency filtering is performed in each formed region. Further, the third detecting means 35 also uses a two-dimensional detector array similar to that of the first embodiment in the same arrangement. The information detected by the second detecting means 35 is further sent to the recognizing means 4, but as in the first embodiment, even if the input image is deformed by shifting, rotating, or scaling, the same object is used. It goes without saying that an image processing apparatus that can be recognized as being present can be configured.

[Embodiment 4] The image processing apparatus of the present embodiment is
This will be described with reference to the diagram showing the configuration of FIG. In the present embodiment, the image input device of the image input means used in the identification optical system 3 of the third embodiment is replaced with an optical writing type and reflection type spatial light modulator, and the entire device is simplified. The configuration is the same as that of the first embodiment except the coordinate conversion optical system 2 and the identification optical system 3. For the sake of simplicity, only the coordinate conversion optical system 2, the identification optical system 3, the recognition system 4 and the parts associated therewith are shown in the drawing.

The coordinate conversion optical system 2 of this embodiment is a second optical input device 21 and second illumination means 22 for inputting the Fourier spectrum information converted by the Fourier conversion optical system 1 into the system. An image input unit, a CGH 23 for superimposing phase information for logarithmic polar coordinate conversion (Logr-θ conversion) on the Fourier spectrum information of the image input into the system by the image input unit, and further, this CGH 23. The second Fourier transform lens 24 for Fourier transforming information is used to transform the object information to be recognized into invariant information for rotation and enlargement / reduction. The difference from the first embodiment is as follows. The other points are the same except that the second detecting means is not necessary to directly write the converted information in the next stage. Further, the identification optical system 3 of the present embodiment includes an image input device 31 which is a third image input means for displaying information processed through the Fourier transform optical system 1 and the coordinate transform optical system 2 in the system. , Substantially parallel coherent light flux and multiplexing means 33 which is a third illuminating means 32 for generating collimated light from various directions and reading out the displayed image, and the read information is superposed. In each area formed by the collimated light in each direction, it comprises a filtering means 34 for different spatial frequency filtering and a third detection means 35, which is Fourier transformed and further coordinate transformed to shift, rotate and scale. Of the information processed into the invariant information for the transformation of the above, the characteristics at a plurality of spatial frequencies are instantly detected, and this information is transmitted to the recognition system 4 in the next stage. It is.

Here, as the substantially parallel coherent light beam which is the illuminating means 32, the light beam 12 used in the Fourier transform optical system 1 is used for simplicity in this embodiment. Specifically, the light flux obtained by bending the light flux transmitted through the beam splitter 12d by the mirror 12i, the mirror 12j, and the mirror 12k is used as the illumination means 32 and is a substantially parallel coherent light flux. The reference numeral 12l is a diaphragm for adjusting the diameter of the light beam, and is not essential. The substantially parallel coherent light flux as the illuminating means 32 is converted into collimated light from various directions by a confocal arrangement refractive index distribution type microlens array 33b ′ and a collimator lens 33a as the multiplexing means 33. It The two-dimensional lens array 33b 'is installed so that the optical axis of each lens is parallel to the optical axis of the collimator lens 33a. The information input from the coordinate conversion optical diameter 2 is superimposed on the collimated light from the various directions by the image input device 31, and the image input device 31 has a writing surface with a second Fourier transform lens. The liquid crystal spatial light modulator 31a of the optical writing type and the reflection type, which is similar to the spatial light modulator 34b of the first embodiment, in which the read surface is provided at the rear focal position of 24 and the read surface is provided at the rear focal position of the collimator lens 33a. And this liquid crystal spatial light modulator 31a
Driver 31b for driving the. Of course, other spatial light modulators can be used as in the case of the first embodiment. The collimated light in each direction in which the information is superposed is reflected by the liquid crystal spatial light modulator 31a and collimator lens 3
3A is reversed, Fourier-transformed, and further, the collimator lens 33a is installed at the rear focal position of the first embodiment.
The same processing is performed by the filtering means 34 including a two-dimensional spatial frequency filter array 34e similar to the above, and different spatial frequency filtering is performed in each region.

As can be seen from the drawings and the description, the identification optical system 3 divides the lens 33a into two parts with respect to the optical axis,
Since one is used as a collimator lens for generating collimated light from various directions, and the other is used as a Fourier transform lens for Fourier transforming each collimated light in which information read by this is superimposed.
The system is compact, but the number of types of spatial frequency filtering that can be handled is halved. Further, the third detecting means 35 also uses a two-dimensional detector array similar to that of the first embodiment in the same arrangement. The information detected by the third detecting means 35 is further sent to the recognition system 4, but it is the same object even if the input image is deformed by shifting, rotating, or scaling like the first embodiment. It goes without saying that an image processing device that can be recognized can be configured.

The image processing apparatus of the present invention has been described above based on some embodiments, but the present invention is not limited to these embodiments and various modifications can be made.

[0035]

As is apparent from the above description, according to the image processing apparatus of the present invention, the input object converted into the invariant information for shifting, rotation and scaling by the Fourier transform optical system and the coordinate transform optical system. The image information of is extracted instantaneously by the identification optical system as a plurality of spatial frequency information that is most effective for distinguishing its features, and further this information is recognized by the neural network which is a recognition system. It is now possible to identify and recognize images that have undergone deformation such as rotation and enlargement / reduction with high accuracy and at high speed.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of an image processing apparatus according to a first embodiment of the present invention.

FIG. 2 is a diagram showing a configuration example of a two-dimensional spatial frequency filter array.

FIG. 3 is a diagram showing a modified example of a combination of a collimator lens and a two-dimensional lens array.

FIG. 4 is a diagram illustrating a main part of a configuration of an image processing apparatus according to a second embodiment.

FIG. 5 is a diagram illustrating a main part of a configuration of an image processing apparatus according to a third embodiment.

FIG. 6 is a diagram illustrating a main part of a configuration of an image processing apparatus according to a fourth exemplary embodiment.

FIG. 7 is a diagram showing a conventional logarithmic polar coordinate conversion optical system.

FIG. 8 is a diagram showing a conventional correlation optical system.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Fourier transform optical system 2 ... Coordinate transform optical system 3 ... Identification optical system 4 ... Recognition system 11 ... Image input device 11a ... Imaging element 11b ... Spatial light modulator 11c ... Driver 12 ... 1st illumination means 12a ... Coherent light source 12b ... beam expander 12b ₁ ... diaphragm lens 12b ₂ ... pinhole 12b ₃ ... collimator lens 12c, 12f, 12g, 12i, 12j, 12k ... mirrors 12d, 12e, 12h, 12m ... beam splitter 12l ... diaphragm 13 ... first Fourier Transforming lens 14 ... Screen 15 ... Two-dimensional imaging device 21 ... Image input device 21a ... Spatial light modulator 21b ... Driver 22 ... Second illuminating means 23 ... CGH (Computer hologram) 24 ... Second Fourier transform lens 25 ... Screen 26 ... Two-dimensional image pickup device 31 ... Image input device 31a Spatial light modulator 31b ... Driver 32 ... Third illuminating means 33 ... Multiplexing means 33a ... Collimator lens 33b ... Two-dimensional lens array 33b '... Gradient index distribution type microlens array 33b "... Gradient index distribution type rod lens array 33c, 33d ... Lens 33e ... Two-dimensional grating 34 ... Filtering means 34a ... Fourth illumination means 34b ... Spatial light modulator 34c ... Driver 34d ... Two-dimensional Fourier transform lens array 34d '... Fourier transform lens 34e ... Two-dimensional spatial frequency filter array 35 ... Third detection means

Claims (5)

[Claims] 1. A first image input means for inputting an image to be recognized into the system, and a first image input means for converting the image information input into the system into Fourier spectrum information. The first Fourier transform lens and this first
Fourier transform optical system including first detecting means for detecting the Fourier spectrum information output to the Fourier transform plane placed at the position of the rear focal point of the Fourier transform lens of, and the detected Fourier spectrum Second image input means for inputting information into the system, and for superimposing phase information for desired coordinate conversion on the Fourier spectrum information of the image input into the system by the second image input means. The computer generated hologram, the second Fourier transform lens for Fourier transforming the superimposed information, and the coordinate transform plane placed at the rear focal point of the second Fourier transform lens. A coordinate conversion optical system including second detection means for detecting coordinate conversion information;
Third image input means for inputting the detected coordinate conversion information into the system, and multiplexing means for multiplexing the coordinate conversion information input into the system by the third image input means. A filtering means for filtering the multiplexed coordinate transformation information in different spatial frequencies in the respective areas, and a third means for detecting the coordinate transformation information in the individual areas subjected to the spatial frequency filtering by the filtering means. An image processing apparatus comprising: an identification optical system including a detection unit; and a recognition system that recognizes an input image based on an output detected by the identification optical system. 2. A system comprising a two-dimensional lens array as the multiplexing means, and a two-dimensional Fourier transform lens array and a two-dimensional spatial frequency filter array arranged on the Fourier transform plane as the filtering means. The image processing apparatus according to claim 1, wherein the image processing apparatus is used. 3. A system comprising an imaging optical system and a two-dimensional diffractive means arranged in the system is used as the multiplexing means, and a two-dimensional Fourier transform lens array and its Fourier transform are used as the filtering means. The image processing apparatus according to claim 1, wherein a system having a two-dimensional spatial frequency filter array arranged on a plane is used. 4. An optical system for irradiating the collimated light from a plurality of different directions with respect to the coordinate conversion information input into the system by the third image inputting means, as the multiplexing means, and the filtering. The image processing apparatus according to claim 1, wherein a system including a Fourier transform lens and a two-dimensional spatial frequency filter array arranged on the Fourier transform surface is used as the means. 5. The image processing apparatus according to claim 4, wherein an optical writing type reflective spatial light modulator serves as both the second detecting means and the third image inputting means. .