JPH075489A - Image processing device and image processing method - Google Patents

Abstract

PURPOSE:To provide the image processing device and image processing method capable of exactly extracting the intrinsic characteristic quantity of an input image at a high speed with simple constitution. CONSTITUTION:This image processing device 101 is composed of a spatial optical modulation element 103 and a uniaxial parallel flat plate optical crystal 102. This spatial optical modulation element 103 is constituted by forming at least a photoconductive layer 106 having a current rectifying property and a liquid crystal layer (optical modulation layer) 112 between two sheets of transparent conductive electrodes 105 and 114 facing each other. The optical crystal 102 is disposed on the photoconductive layer 106 side of the spatial optical modulation element 103. The angle formed by the optical axis of the optical crystal 102 and the normal of the crystal surface is expressed by equation 0<xsi<90.

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 and an image processing method for processing and recognizing images such as characters and two-dimensional patterns.

[0002]

2. Description of the Related Art In recent years, image processing using a neural network has been studied, and in particular, an optical neural network is expected to be realized as an effective means for processing two-dimensional image information at high speed (Masatoshi Ishikawa, Optics, No. 19
Vol. 11, No. 755-761 (1990)). This uses an optical signal as an information medium, and pays attention to the spatial parallelism and high speed of light. Above all, an image processing device using a spatial light modulator is important as a key device when configuring an optical neural network.

FIG. 15 shows a sectional view of a spatial light modulator used in a conventional image processing apparatus. The spatial light modulator 1501 includes two substrates 1502 and 150 facing each other.
9, transparent conductive electrodes 1503 and 1508, photoconductive layer 1
504, the dielectric reflection layer 1506, and the light modulation layer 1507 are sandwiched between them. Then, since the electric resistance of the photoconductive layer 1504 is lowered and the voltage applied to the light modulation layer 1507 is increased in the portion where the input light 1510 enters, the read light 1511 can be modulated.

[0004]

By the way, in order to improve the recognition ability of the optical neural network, it is important to quickly and accurately extract the characteristic amount peculiar to the image.

However, the conventional spatial light modulator has only a function as an input device or executes a simple logical operation, and it is not possible to extract a characteristic amount peculiar to an image. It was possible.

Further, in order to extract the feature amount of the input image, the feature amount must be extracted by a computer after photoelectrically converting the input image and accumulating it in the image memory, which requires a huge amount of time for processing. However, there was a problem that the configuration became complicated.

In order to solve the above-mentioned problems of the prior art, the present invention provides an image processing apparatus and an image processing method which have a simple structure and can extract the characteristic amount of the input image at high speed and accurately. With the goal.

[0008]

In order to achieve the above object, an image processing apparatus according to the present invention comprises a spatial light modulator and
An image processing apparatus comprising at least a uniaxial parallel plate optical crystal, wherein the spatial light modulation element has a rectifying photoconductive layer and a light modulation layer between two opposing transparent conductive electrodes. At least, the optical crystal is provided on the photoconductive layer side of the spatial light modulation element, and the angle ξ formed by the optical axis of the optical crystal and a normal line to the crystal surface is represented by the (Equation 1). Is characterized by.

Further, in the above structure, it is preferable that the optical crystal is divided into a plurality of regions, and the direction in which the optical axis is projected on the crystal surface is different for each of the plurality of regions. In this case, the optical crystal is divided into at least four regions, and the angle at which the projection directions intersect is approximately 45.
It is preferably one selected from degrees, approximately 90 degrees and approximately 135 degrees.

Further, in the above structure, it is preferable that the photoconductive layer and the light modulation layer of the spatial light modulation element are electrically connected to each other through the metal reflection film separated into minute shapes. Further, in the above structure, the light modulation layer is preferably made of a ferroelectric liquid crystal.

In the above structure, the spatial light modulator has a threshold value for the incident light quantity, the light output for the incident light quantity below the threshold value is substantially zero, and the light output for the incident light quantity exceeding the threshold value is It is preferably substantially constant without depending on the amount of incident light.

The image processing method according to the present invention is an image processing apparatus comprising at least a spatial light modulation element and a uniaxial parallel plate optical crystal, wherein the spatial light modulation elements are two facing each other. At least a photoconductive layer having a rectifying property and a light modulation layer between the transparent conductive electrodes of, the optical crystal is provided on the photoconductive layer side of the spatial light modulation element, the optical axis of the optical crystal An image composed of a two-dimensional matrix is vertically incident as a monochromatic plane wave on the image processing apparatus whose angle ξ formed by the normal to the crystal surface is expressed by the above (Formula 1), and an extraordinary ray generated from the first matrix is detected. It is characterized in that the ordinary ray generated from the second matrix is superposed on the ray emitting side of the optical crystal.

Further, in the constitution of the method of the present invention,
The first matrix and the second matrix are preferably adjacent to each other. In the configuration of the method of the present invention, the amount of ordinary rays and the amount of extraordinary rays are substantially equal,
The threshold value is larger than the light amount of the ordinary ray and the extraordinary ray,
Further, it is preferably smaller than the sum of the light amounts of the ordinary ray and the extraordinary ray.

Further, in the constitution of the method of the present invention,
It is preferable that the monochromatic plane wave is linearly polarized light, and the polarization direction intersects the projection direction at about 45 degrees. In the configuration of the method of the present invention, it is preferable that the output from the image processing device is processed by a neural network to recognize the image.

[0015]

The spatial light modulation element constituting the image processing apparatus of the present invention has a threshold value for the incident light quantity, and the optical output for the incident light quantity below the threshold value is substantially zero. Further, the light output with respect to the incident light amount exceeding the threshold value is substantially constant without depending on the incident light amount. A uniaxial parallel plate optical crystal is provided on the light incident side of this spatial light modulator, and the angle ξ formed by the optical axis of the optical crystal and the normal line to the crystal surface satisfies the above (Equation 1).

Consider a case where a monochromatic plane wave representing an image formed by a two-dimensional matrix is vertically incident on the image processing apparatus. The incident plane wave is separated into an ordinary ray and an extraordinary ray in the optical crystal and propagates in different directions.
The ordinary ray propagates in the direction normal to the crystal surface, and the extraordinary ray propagates in a specific direction determined by the angle ξ that the optical axis makes with the normal to the crystal surface. At this time, the thickness of the optical crystal and the size of the matrix of the image are determined so that the extraordinary ray generated from the first matrix and the ordinary ray generated from the second matrix overlap each other on the light emitting side of the optical crystal. It is possible to In addition, by utilizing the fact that the direction in which the optical axis is projected on the crystal surface (hereinafter referred to as the projection direction) and the projection direction of the ray direction of the extraordinary ray coincide, the first
It is also possible to arbitrarily determine the direction of the second matrix with respect to the matrix. Further, by making the incident plane wave unpolarized, or by making the incident plane wave linearly polarized and intersecting the polarization direction with the projection direction of the optical axis at 45 degrees, the light quantities I _o and I _e of the ordinary ray and the extraordinary ray are substantially equal. However, the threshold value α of the spatial light modulator can be set to satisfy the following (Equation 2).

[0017]

[Equation 2]

The principle of line segment extraction by the image processing apparatus of the present invention will be described below. By setting the ordinary ray and the extraordinary ray of the vertically adjacent matrix to overlap each other on the light emitting side of the optical crystal, the image processing apparatus of the present invention can extract the vertical line segment included in the image. . Two-dimensional matrix (matrix size is n ×
The matrix is located in row i and column j of the constructed image from n) and x (i, j), the value of the matrix x _ij
(1 or 0). In addition, the light quantity I _ij of the incident plane wave representing the image satisfies the following ( _Equation 3).

[0019]

[Equation 3]

[0020] In equation _(3), I oij, I _eij is the amount of each ordinary ray, the extraordinary ray, satisfies the equation (2). If the extraordinary ray from x (i, j) overlaps with the ordinary ray of x (i + 1, j) on the exit side of the optical crystal, the image emitted from the optical crystal has a matrix size n ×
If it is a two-dimensional (n + 1) -dimensional light pattern and the light quantity of the matrix located in the i-th row and the j-th column is y _ij , y _ij can be expressed by the following ( _Equation 4).

[0021]

[Equation 4]

That is, the two-dimensional light pattern represented by (Equation 4) is input to the spatial light modulator. Therefore, the output of the spatial light modulator is the matrix size n
X (n + 1) two-dimensional pattern, whose output is z _ij
Then, z _ij can be expressed by the following ( _Equation 5).

[0023]

[Equation 5]

From the above (Equation 2), (Equation 3), (Equation 4),
z _ij = 1 when the following (Equation 6) is satisfied.

[0025]

[Equation 6]

That is, this spatial light modulator outputs only when there is a vertical line segment in the input image (when the vertically adjacent matrixes are 1 at the same time). Similarly, by adjusting the projection direction of the optical axis and the thickness of the optical crystal, the horizontal line segment, the left diagonal line segment ("/"), and the right diagonal line segment ("\") can be extracted. At this time, for example, in order to extract a horizontal line segment, it may be set so as to satisfy the following (Equation 7).

[0027]

[Equation 7]

Similarly, in order to extract the left diagonal line segment and the right diagonal line segment, the following equations (8) and (9) may be set, respectively.

[0029]

[Equation 8]

[0030]

[Equation 9]

As described above, according to the image processing apparatus of the present invention, it is possible to selectively extract a specific line segment from the input image. Next, a method of recognizing an image by processing the output of the image processing apparatus of the present invention with a neural network will be described.

It is known that a neural network constructed by imitating the input / output operation of the cerebral nervous system of a living body can exhibit excellent ability for pattern recognition of images and the like. In order to improve the recognition ability, it is indispensable to extract the feature amount that is not easily affected by the pattern deformation. The image processing apparatus of the present invention can extract a feature amount, which is a specific line segment, that is not easily affected by pattern deformation, from an input image. Therefore, if the output of the image processing apparatus of the present invention is used as the input signal of the neural network, a high recognition function can be realized.

[0033]

EXAMPLES The present invention will be described in more detail below with reference to examples. (Embodiment 1) FIG. 1 is a sectional view showing an embodiment of an image processing apparatus according to the present invention. The image processing apparatus 101 includes a spatial light modulator 103 and a uniaxial parallel plate optical crystal (hereinafter referred to as an optical crystal) 102.

The spatial light modulator 103 has the following structure. That is, on the transparent insulating substrate 104 (eg, glass), the transparent conductive electrode 105 (eg, ITO (indium-tin oxide), ZnO, Sn) is formed.
And O _2, etc.), a photoconductive layer 106 having a rectifying property (e.g., an amorphous silicon layer having a diode structure; p layer 1
07, i layer 108, and n layer 109) are sequentially stacked. Further, on the n-layer 109, a separated minute metal reflection film 110 (for example, Al, Ti,
A metal such as Cr or Ag, or a laminate of two or more kinds of metals) and an alignment film 111 for aligning a ferroelectric liquid crystal
(For example, a polymer thin film such as polyimide) is sequentially arranged. In addition, the transparent conductive electrode 114 (eg, I) is also formed on the other transparent insulating substrate 115 (eg, glass).
TO (indium-tin oxide), ZnO, SnO _2, etc. is formed, and an alignment film 113 (for example, a polymer thin film such as polyimide) is applied thereon. And
A liquid crystal layer 112 is formed between both substrates.

The material used for the photoconductive layer 106 is
For example, CdS, CdTe, CdSe, ZnS, ZnS
e, GaAs, GaN, GaP, GaAlAs, InP
Such as compound semiconductor, amorphous such as Se, SeTe, AsSe
Quality semiconductors, Si, Ge, Si _1-xC_x, Si_1-xG
e_x, Ge_1-xC_xPolycrystalline or amorphous such as (0 <x <1)
Quality semiconductor, and (1) Phthalocyanine pigment (abbreviated as Pc)
), For example, metal-free Pc, XPc (X = Cu, Ni,
Co, TiO, Mg, Si (OH)₂Etc.), AlCl
PcCl, TiOClPcCl, InClPcCl, I
(2) Monoazo color such as nClPc, InBrPcBr
Azo dyes such as elementary and disazo dyes, (3) penenylene acid
Penylene-based pigments such as anhydrides and penenylene imides,
(4) indigoid dye, (5) quinacridone pigment,
(6) Polycyclic quinones such as anthraquinones and pyrenequinones
Non-type, (7) Cyanine dye, (8) Xanthene dye,
(9) Charge transfer complex such as PVK / TNF, (10) Bi
Formed from Lilium Salt Dye and Polycarbonate Resin
Eutectic complexes, (11) azurenium salt compounds and other organic semiconductors
I have a body.

Amorphous Si, Ge, Si
_{1-x Cx} , Si _1-x Ge _x , Ge _1-x C _x (hereinafter a-
Si, a-Ge, a-Si _1-x C _x , a-Si _1-x Ge
_x , a-Ge _1-x C _x is abbreviated) as the photoconductive layer 106.
When used as, hydrogen or a halogen element may be included, and oxygen or nitrogen may be included in order to reduce the dielectric constant or increase the resistivity. P for controlling the resistivity
An element such as B, Al, or Ga that is a type impurity, or an element such as P, As, or Sb that is an n-type impurity may be added. In this way, the impurity-added amorphous materials are stacked to form junctions such as p / n, p / i, i / n, and p / i / n, and a depletion layer is formed in the photoconductive layer 106. It is also possible to control the dielectric constant and dark resistance or operating voltage polarity. Also,
In addition to such an amorphous material, two or more kinds of the above materials may be stacked to form a heterojunction and a depletion layer may be formed in the photoconductive layer 106. The thickness of the photoconductive layer 106 is preferably in the range of 0.1 to 10 μm.

The spatial light modulator 103 is driven by applying a pulse voltage between the transparent conductive electrodes 105 and 114. One cycle of this pulse voltage is composed of an erase pulse that is forward biased to the rectifying photoconductive layer 106 and a write pulse that is reverse biased to the photoconductive layer 106. Then, while the write pulse is being applied, the photoconductive layer 1
When the writing light is not input to 06, the liquid crystal layer 11
No. 2 does not switch and remains in the initial state. this is,
This is because the dark current of the photoconductive layer 106 is very small. On the other hand, while the write pulse is being applied, the photoconductive layer 1
When the writing light is input to 06, a photocurrent that is approximately proportional to the incident light intensity is generated in the photoconductive layer 106 and is collected in the metal reflection film 110. Since the liquid crystal layer 112 has a threshold function, when the sum of the collected photocurrents exceeds a certain threshold value, the liquid crystal layer 112 switches to modulate the read light 117. That is, the spatial light modulator 103 has a binarizing function for the amount of input light. still,
The liquid crystal layer 112 is forcibly initialized when the erase pulse is applied.

Next, a method of manufacturing the spatial light modulator having the above structure will be described with reference to FIG. First, 40m
On a glass substrate 104 of m × 40 mm × 0.3 mm, ITO is formed into a film having a film thickness of 1000 angstrom by a sputter vapor deposition method to form a transparent conductive electrode 105.
Then, hydrogenated amorphous silicon (a-Si: H) having a p / i / n diode structure is formed by a plasma CVD method.
The films are stacked with a thickness of 2 μm to form the photoconductive layer 106. Here, B (boron) is used as an impurity in the p layer 107.
Is doped at 400 ppm, and the n-layer 109
Is doped with P (phosphorus) by 40 ppm.
Further, the p-layer 107 has a film thickness of 1000 angstroms,
The thickness of the i layer 108 is 17,000 angstroms, and the thickness of the n layer 109 is 2000 angstroms. Then, Cr is deposited on the photoconductive layer 106 by a vacuum deposition method.
(Chromium) is formed to a film thickness of 2000 angstrom, and the metal reflection film 110 is patterned by photolithography. That is, the size of each pixel is 20μ
m × 20 μm, pitch 5 μm, 1000 × 10
A metal reflection film 110 of 00 pixels, which is separated from each other, is formed. Then, a polyamic acid is applied thereon by a spin coating method, and thermosetting is applied to the polyimide alignment film 1.
11 is formed. At this time, the film thickness of the polyimide alignment film 111 is 100 angstrom. The orientation treatment (rubbing treatment) is performed by rubbing the surface in a certain direction with a nylon cloth.

Similarly, a transparent conductive electrode 114 made of ITO and a polyimide alignment film 113 are sequentially formed on the other glass substrate 115 and subjected to alignment treatment. Then, beads having a diameter of 1 μm are distributed on the glass substrate 115, and the glass substrate 104 is bonded to form a gap of 1 μm between the two substrates. Then, the liquid crystal layer 112 having a thickness of 1 μm is formed between both substrates by injecting a ferroelectric liquid crystal into the gap and performing a heat treatment. From the above, the spatial light modulator 103 shown in FIG. 1 is obtained.

In the image processing apparatus 101 of the present invention,
The optical crystal 1 is provided on the photoconductive layer 106 side of the spatial light modulator 103.
02 is provided. As the optical crystal 102,
It may be a uniaxial crystal or a negative uniaxial crystal. Positive uniaxial bond
As the crystal, for example, quartz (SiO₂), Rutile (Ti
O₂), Magnesium fluoride (MgF₂) Etc.
be able to. Also, as a negative uniaxial crystal, for example,
For example, calcite (CaCO ₃), Sodium nitrate (NaNo
₃), Aluminum oxide (Al₂O₃) Etc.
You can

Next, the operation of the optical crystal 102 will be described. Generally, a plane wave incident on an anisotropic optical crystal is separated into an ordinary ray and an extraordinary ray and propagates in the crystal.
Therefore, two light rays are observed on the light emitting side of the crystal. The propagation of light waves in such an anisotropic optical crystal can be explained using a wavefront normal ellipsoid. FIG. 2 shows a wavefront normal ellipsoid for determining the propagation direction of a monochromatic plane wave incident on a positive uniaxial optical crystal.
If the z axis is taken in the direction of the optical axis, the x axis and the y axis can be taken arbitrarily, but for simplicity, the plane of polarization of the extraordinary ray is zx.
Face. Assuming that the wavefront normal direction of the extraordinary ray is s, the electric field vector is E, and the electric displacement vector is D, the following (Equation 10) is established.

[0042]

[Equation 10]

In (Equation 10), ψ and θ are angles formed by D and E with the z-axis, respectively. From (Equation 10), the tangent of ψ can be expressed by the following (Equation 11).

[0044]

[Equation 11]

In (Equation 11), n _o and n _e are refractive indices for ordinary rays and extraordinary rays, respectively. When the ray direction of the extraordinary ray is t and the angle formed with the wavefront normal direction s is δ, the tangent of δ can be expressed by the following (Equation 12).

[0046]

[Equation 12]

Since the ray direction t of the ordinary ray coincides with the wavefront normal direction s, δ is equal to the difference between the ray directions of the ordinary ray and the extraordinary ray. From (Equation 12), the maximum value of | tanδ |
It is given by the following equation (13) when the tanψ = n _o / n _e.

[0048]

[Equation 13]

A similar analysis holds for a negative uniaxial optical crystal. FIG. 3 shows a wavefront normal ellipsoid in the case of a negative uniaxial optical crystal. Here, the optical axis is taken as the x-axis and the plane of polarization of extraordinary rays is taken as the zx plane. Then, also in this case, the above (Equation 12) and (Equation 13) are satisfied as they are, and the ordinary ray and the extraordinary ray are separated and propagate in the crystal.

Based on the above, the parallel plate positive uniaxial optical crystal 1 used in the image processing apparatus 101 of the first embodiment.
02, the propagation of a light wave when a monochromatic plane wave is vertically incident will be described.

FIG. 4 shows a wavefront normal ellipsoid. The wavefront normal direction s of the extraordinary ray coincides with the normal direction of the crystal surface Σ, but the ray direction t is deviated from the wavefront normal direction s by an angle δ. The tangent of δ is expressed by the above (Equation 12), and the angle ξ (0 <ξ = 90−ψ <90) formed by the optical axis and the normal to the crystal surface is formed.
Can be expressed by the following (Equation 14).

[0052]

[Equation 14]

On the other hand, the ray direction t of the ordinary ray in the crystal coincides with the normal direction of the crystal surface Σ. Therefore, assuming that the thickness of the crystal is d, the ordinary ray and the extraordinary ray are deviated on the light emitting side of the crystal, and the deviation width Δ is as follows.
It can be described in 5).

[0054]

[Equation 15]

Here, the functions of the optical crystal 102 are summarized as follows. (1) A vertically incident plane wave propagates in the optical crystal 102 while being separated into an ordinary ray and an extraordinary ray. (2) The deviation width Δ between the ordinary ray and the extraordinary ray on the back surface of the crystal can be adjusted by the angle ξ formed by the optical axis and the normal to the crystal surface and the thickness d of the crystal. (3) The deviation direction of the extraordinary ray can be arbitrarily set by utilizing the fact that the direction of the extraordinary ray projected on the crystal surface and the projection direction of the optical axis coincide with each other.

Next, an image processing method for extracting vertical line segments from an image using the image processing apparatus 101 of the present invention will be described. FIG. 5 is a plan view of an optical crystal used in an image processing apparatus for extracting a vertical line segment, as viewed from the light incident side. A monochromatic plane wave representing an image is vertically incident on the optical crystal 501. Therefore, the wavefront normal direction of the incident plane wave is perpendicular to the paper surface. At this time, the incident plane wave is linearly polarized, and its polarization direction 503 is set to intersect the projection direction 502 of the optical axis on the crystal surface at 45 degrees. As a result, the light quantity I _{o of the} ordinary ray and the light quantity I _{e of the} extraordinary ray generated in the crystal can be made substantially equal.

Now assume that the incident plane wave represents the input image shown in FIG. 6 (a). The coordinate system at this time is defined as described in the above (action). That is, a two-dimensional matrix (the matrix size is 8 ×
The matrix located at the i-th row and the j-th column of the image composed of 8) is x (i, j), and the value of the matrix is x _ij
(1 or 0). Further, the light quantity I _ij of the incident plane wave representing the image satisfies the above-mentioned ( _Equation 2) and (Equation 3). In addition, by making the deviation width Δ between the ordinary ray and the extraordinary ray expressed by the above (Equation 15) substantially equal to the size of one matrix, the image emitted from the optical crystal 501 is a two-dimensional matrix of size 8 × 9. If y _ij is the light quantity of the matrix located at the i-th row and the j-th column, y _ij can be expressed by the above ( _Equation 4).

Therefore, the light emitted from the optical crystal 501 (that is, the incident pattern on the spatial light modulator 103) is as shown in FIG. 6B. Where matrix 60
Reference numeral 1 indicates an extraordinary ray, matrix 602 indicates an ordinary ray, and matrix 603 indicates a portion where the ordinary ray and the extraordinary ray overlap. Since the light quantities of the ordinary ray and the extraordinary ray satisfy the above (Equation 2), the output Z of the spatial light modulator 103 is as shown in FIG. 6C from the above (Equation 4) and (Equation 5). Figure 6
As is apparent by comparing (c) with FIG. 6A which is the input image, the output of the image processing apparatus 101 of the present invention is the result of selectively extracting only the vertical line segment from the input image.

(Embodiment 2) FIG. 7 is a plan view showing an optical crystal used in an image processing apparatus capable of selectively extracting line segments in four directions. This optical crystal 701 has four regions (702 to 702).
705), and the projection direction of the optical axis in each region onto the crystal surface is different for each region. That is, the angles formed by the respective projection directions are 45 degrees and 90 degrees.
Or 135 degrees. Further, the shift width Δ between the regions 702 and 703 is substantially equal to the size of one matrix forming the input image, and the shift width Δ ′ between the regions 704 and 705 and the following (Equation 16) are satisfied. Is set.

[0060]

[Equation 16]

By making the incident light non-polarized,
The light quantity of the ordinary ray and the light quantity of the extraordinary ray in each region are made substantially equal. The structure of the spatial light modulator used in the image processing apparatus of the second embodiment is the same as that of the spatial light modulator 103 used in the first embodiment.

The four areas of the image processing apparatus are shown in FIG.
When a monochromatic plane wave representing the image shown in (a) was simultaneously input to each area, the output shown in FIG. 8 was obtained. From this, it was confirmed that line segments in four directions can be selectively extracted from the input image by using the image processing apparatus of the second embodiment.

(Embodiment 3) FIG. 9 is a sectional view showing another embodiment of the image processing apparatus according to the present invention. The basic structure and operation of the image processing apparatus 901 of the third embodiment are the same as those of the image processing apparatus 101 of FIG. 1, except that a dielectric reflecting film 907 is formed on the entire surface instead of the metal reflecting film 110. The spatial light modulator 903 is provided. Also, the optical crystal 9
02 has the same configuration and arrangement as the optical crystal 701 used in the second embodiment.

In the four areas of the image processing apparatus 901,
When a monochromatic plane wave representing the image shown in FIG. 6 (a) was simultaneously input to each region, the output shown in FIG. 8 was obtained. Accordingly, if the image processing apparatus 901 of the third embodiment is used,
It was confirmed that line segments in four directions can be selectively extracted from the input image.

(Embodiment 4) FIG. 10 is a sectional view showing still another embodiment of the image processing apparatus according to the present invention. The basic structure and operation of the image processing apparatus 1001 according to the fourth embodiment are similar to those of the image processing apparatus 101 of FIG. 1, but the optical crystal 1002 is used as a substrate on the incident side of the spatial light modulator 1003. It's different. Optical crystal 1002
The configuration and arrangement are the same as those of the optical crystal 701 used in Example 2.

When a monochromatic plane wave representing the image of FIG. 6A was simultaneously input to each of the four areas of the image processing apparatus 1001, the output shown in FIG. 8 was obtained. Thus, if the image processing apparatus 1001 of the fourth embodiment is used,
It was confirmed that line segments in four directions can be selectively extracted from the input image.

(Embodiment 5) FIG. 11 is a block diagram of an image recognition apparatus constructed by using the image processing apparatus used in Embodiment 2 based on the image processing method of the present invention. The image recognition device 1101 according to the fifth embodiment includes a laser 1102, a tablet 1103, a beam expander 1104, a liquid crystal television 1105, an image processing device 1111, a surface light source 1106, a polarizer 1107, a beam splitter 1108, and an analyzer 11.
09, a CCD camera 1110, and a computer 1112. In the image recognition device 1101, the beam from the laser 1102 is expanded by the beam expander 1104 and input to the liquid crystal television 1105. The liquid crystal television 1105 displays four images input from the tablet 1103. Image processing device 111
1 selectively extracts line segments in four directions from the input image. The extracted result is that the readout light from the surface light source 1106 is input to the ferroelectric liquid crystal of the image processing apparatus 1111 via the polarizer 1107, and the reflected light is reflected by the beam splitter 110.
Can be obtained by passing through the analyzer 1109. The output of the image processing device 1111 is captured by the CCD camera 1110, and the computer 1
Sent to 112. The computer 1112 processes an image pickup result by a neural network and recognizes an image.

Next, the neural network model used in the image recognition apparatus of the fifth embodiment will be described. Figure 1
Figure 2 shows a schematic diagram of the neural network model. The model used has four neuron layers in which two-dimensionally arranged neurons (nerve cells) that perform input / output operations similar to those of human cerebral nerve cells are arranged. This neural network model is used to calculate "vertical", "horizontal", "vertical" from an input image given by a two-dimensional bit image as shown in FIG.
The image is recognized by extracting line segment information indicating at what position and in what length the line segment in four directions of "left diagonal" and "right diagonal" exists. Therefore, by applying this neural network model to, for example, character recognition, handwritten characters can be recognized.

Next, the operation of the neural network will be described. The input layer 1201 is 8 shown in FIG.
It is for inputting a two-dimensional bit image having 64 data of 8 × 8, and is constituted by 64 neurons of 8 × 8. The neurons of the second layer 1202 have four regions 1203 to 120, each of which has 8 × 8 as one unit.
It is divided into 6 and has a total of 8 × 8 × 4 256 neurons. Each area 1203 of the second layer 1202
Reference numerals 1206 to 1206 are specially coupled to the input layer (first layer) 1201, and each region corresponds to a line segment in four directions forming a two-dimensional bit image. That is, each neuron in the region 1203 has an input layer (first layer) 1201
The "vertical direction" line segment can be extracted by connecting the corresponding neuron (having the same two-dimensional coordinates) and the adjacent neuron thereamong.
Similarly, each neuron in the region 1204 has a corresponding neuron and a neuron adjacent to the left of the corresponding neuron in the region 1205.
Each of the neurons of “1” is connected to the corresponding neuron and its adjacent diagonally upper right, and each of the neurons in the region 1206 is connected to the corresponding neuron and the adjacent neuron adjacent to its left diagonal “laterally”,
It is possible to extract line segments in the “left diagonal direction” and the “right diagonal direction”. These bonds are respectively expressed by the above (Equation 4),
It corresponds to (Equation 7), (Equation 8), and (Equation 9), and can be realized by the optical crystal 701 used in the second embodiment.

Each area of the second layer 1202 is divided into nine small areas. That is, it is divided into three regions in the row and column directions. Since this small area corresponds to dividing the two-dimensional bit image into nine areas, for example, the output of the neuron in the area 1203 means that a vertical line segment exists, and the area 1203 The existence position of the vertical line segment can be confirmed depending on which of the small regions of the neuron outputs. Furthermore, all neurons in each subregion are
It is connected to one corresponding neuron in the third layer 1207. Therefore, each neuron in the third layer 1207 has a second
The direction and position of the corresponding line segment differ depending on which small region of the layer 1202 the neuron is connected to.
That is, the output state of each neuron of the third layer 1207 is a line segment that constitutes a two-dimensional bit pattern, and indicates in what direction and in what part of the line segment the line segment exists. Represents the unique features of.

Each neuron in the third layer 1207 is connected to all neurons in the fourth layer 1208, and only the neuron having the largest total sum of input signals is output in the fourth layer 1208. When recognizing 26 letters of the alphabet, the fourth layer 1208 has 26 neurons, and each neuron corresponds to one alphabet. The alphabet corresponding to the output neuron is the recognition result. In addition, learning is performed in the third layer 1207 and the fourth layer 1
The connection of 208 is performed using the orthogonal learning method, and all other synaptic weights are fixed to the same value.

Next, a specific recognition process will be described with reference to FIGS. 11 and 13. FIG. 13 (a),
(A ′) is a handwritten character input to the tablet 1103. This character is converted into an 8 × 8 matrix pattern as shown in (b) and (b ′), and only four characters are displayed on the liquid crystal television 1105. The laser light expanded by the beam expander 1104 enters the liquid crystal television 1105 and is vertically input to the image processing device 1111 as a plane wave representing an image. Image processing device 1111
Output is 4 as described in the first and second embodiments.
13 is a result of selectively extracting line segments in the direction, and FIG.
It becomes like (c) and (c '). This is the output of the middle tier. This image is captured by the CCD camera 1110, and the second
The sum of each small area of the layer is the third layer 1207 (see FIG. 1).
2) (see FIGS. 13 (d) and 13 (d ')). It is the length of the line segment to which the diameter of the black circle is added. Third layer 120
The final recognition is performed by connecting each neuron of No. 7 to all the neurons of the fourth layer 1208 (see FIG. 12). The operations of the third layer 1207 and the fourth layer 1208 are as follows. It is realized by the computer 1112.

By learning the 26 letters of the alphabet in advance, the image recognition apparatus of the fifth embodiment can be used as shown in FIG.
It was confirmed that the handwritten characters shown in can be correctly recognized and that it has high recognition ability.

Since the image processing apparatus used in the above-mentioned Examples 1 to 5 uses the ferroelectric liquid crystal capable of high-speed response as the light modulation layer, processing of 3000 patterns per second is possible.

[0075]

As described above, according to the image processing apparatus of the present invention, it is possible to quickly and accurately extract the line segments that make up the input image with a simple structure.

Further, according to the image processing method of the present invention, various line segments included in a character pattern can be extracted at high speed. Therefore, image processing with high recognition ability can be performed by combining with a neural network. You can

[Brief description of drawings]

FIG. 1 is a sectional view showing an embodiment of an image processing apparatus according to the present invention.

FIG. 2 is an explanatory diagram showing a wavefront normal ellipsoid for determining the propagation direction of an extraordinary ray in a positive uniaxial optical crystal.

FIG. 3 is an explanatory diagram showing a wavefront normal ellipsoid for determining the propagation direction of an extraordinary ray in a negative uniaxial optical crystal.

FIG. 4 is an explanatory diagram showing a wavefront normal ellipsoid for determining a ray direction of an extraordinary ray of a monochromatic plane wave vertically incident on a positive uniaxial optical crystal.

FIG. 5 is a plan view showing an optical crystal used in an embodiment of the image processing apparatus according to the present invention.

6A is an explanatory view showing an input image to the image processing apparatus according to the present invention, FIG. 6B is an explanatory view showing an output image of an optical crystal, and FIG. 6C is an image processing apparatus according to the present invention. It is explanatory drawing which shows an output image.

FIG. 7 is a plan view showing an optical crystal used in another embodiment of the image processing apparatus according to the present invention.

FIG. 8 is an explanatory diagram showing an output image of the image processing apparatus according to the present invention.

FIG. 9 is a cross-sectional view showing another embodiment of the image processing apparatus according to the present invention.

FIG. 10 is a sectional view showing still another embodiment of the image processing apparatus according to the present invention.

FIG. 11 is a configuration diagram of an image recognition device based on an image processing method according to the present invention.

FIG. 12 is a schematic diagram of a neural network model.

FIG. 13 is an explanatory diagram showing a recognition process by the image recognition device.

FIG. 14 is an explanatory diagram showing an example of characters that can be correctly recognized by the image recognition device according to the present invention.

FIG. 15 is a cross-sectional view showing a conventional spatial light modulator.

[Explanation of symbols]

 101 Image Processing Device 102 Uniaxial Parallel Plate Optical Crystal 103 Spatial Light Modulating Element 104, 115 Transparent Insulating Substrate 105, 114 Transparent Conductive Electrode 106 Photoconductive Layer 107 p Layer 108 i Layer 109 n Layer 110 Metal Reflective Film 111, 113 Alignment film 112 Liquid crystal layer

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kunihito Ogawa 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd.

Claims (11)

[Claims] 1. An image processing device comprising at least a spatial light modulator and a uniaxial parallel plate optical crystal, wherein the spatial light modulator rectifies between two transparent conductive electrodes facing each other. At least a photoconductive layer and a light modulation layer having optical properties, the optical crystal is provided on the photoconductive layer side of the spatial light modulation element, and an angle ξ formed by the optical axis of the optical crystal and a normal line to the crystal surface. Is represented by the following (Equation 1). [Equation 1] 2. The image processing apparatus according to claim 1, wherein the optical crystal is divided into a plurality of regions, and the direction in which the optical axis is projected on the crystal surface is different for each of the plurality of regions. 3. The optical crystal has at least four divided regions, and the angle at which the projection directions intersect is approximately 45.
3. The image processing apparatus according to claim 2, wherein the image processing apparatus is one selected from degrees, approximately 90 degrees, and approximately 135 degrees. 4. The image according to claim 1, 2 or 3, wherein the photoconductive layer and the light modulation layer of the spatial light modulation element are electrically connected via a metal reflection film separated into minute shapes. Processing equipment. 5. The image processing device according to claim 1, wherein the light modulation layer is made of a ferroelectric liquid crystal. 6. The spatial light modulator has a threshold value for the incident light quantity, the light output for the incident light quantity below the threshold value is substantially zero, and the light output for the incident light quantity exceeding the threshold value depends on the incident light quantity. Claim 1 which is substantially constant without
The image processing device according to 2, 3, 4 or 5. 7. An image processing apparatus comprising at least a spatial light modulator and a uniaxial parallel plate optical crystal, wherein the spatial light modulator rectifies between two transparent conductive electrodes facing each other. At least a photoconductive layer and a light modulation layer having optical properties, the optical crystal is provided on the photoconductive layer side of the spatial light modulation element, and an angle ξ formed by the optical axis of the optical crystal and a normal line to the crystal surface. Is incident on the image processing device represented by the above (Formula 1) as an image composed of a two-dimensional matrix vertically as a monochromatic plane wave, and an extraordinary ray generated from the first matrix and an ordinary ray generated from the second matrix. And an image processing method in which the above are overlapped on the light emitting side of the optical crystal. 8. The image processing method according to claim 7, wherein the first matrix and the second matrix are adjacent to each other. 9. The light amount of an ordinary ray and the light amount of an extraordinary ray are substantially equal to each other, and the threshold value is larger than the light amounts of the ordinary ray and the extraordinary ray and smaller than the sum of the light amounts of the ordinary ray and the extraordinary ray. Item 9. The image processing method according to Item 7 or 8. 10. The image processing method according to claim 7, 8 or 9, wherein the monochromatic plane wave is linearly polarized light, and the polarization direction intersects the projection direction at about 45 degrees. 11. The image processing method according to claim 7, wherein the output from the image processing apparatus is processed by a neural network to recognize an image.