JP2900209B2 - Optical pattern recognition method and optical pattern recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the field of optical information processing and optical measurement, in which a two-dimensional image obtained from an imaging device such as a CCD camera is subjected to a correlation process using coherent light to thereby perform pattern recognition. And a device for automatically performing measurement.

[0002]

2. Description of the Related Art Conventionally, many studies have been made on a pattern recognition method or a pattern recognition device by performing a correlation process using coherent light. In many of these methods, an input image is optically Fourier-transformed, the Fourier-transformed image is filtered by an optical filter created by a predetermined method, and then optically Fourier-transformed again. This is an optical Fourier transform correlation method that attempts to correlate with image information in a filter. The optical Fourier transform correlation method includes a matched filter type optical correlator that filters a Fourier transform image of a correlated image using a Fourier transform hologram of a reference image as the optical filter, and a Fourier transform of the correlated image and the reference image simultaneously. There is a joint transform optical correlator that creates a joint Fourier transform hologram by transforming it to produce a joint Fourier transform hologram and optically Fourier transforms it again. The matched filter type optical correlator is suitable for performing precise optical pattern recognition because alignment is difficult but the SN ratio is large. The joint conversion optical correlator is suitable for real-time optical pattern recognition because the adjustment of the optical system is extremely easy and the correlation processing can be performed in one process.

Further, recently, by binarizing an optical filter, the SN of the conventional optical Fourier transform correlation method has been improved.
The ratio has improved significantly. Furthermore, as a recording medium for an optical filter, a liquid crystal television, an optical writing type liquid crystal spatial light modulator, various photorefractive optical crystals, and the like have been used in place of a conventional silver halide photographic dry plate. Research for optical pattern recognition has been actively conducted.

[0004]

However, irrespective of the conventional optical Fourier transform correlation method, in other optical pattern recognition methods, the correlated image does not correspond to the reference image recorded in the optical filter. If the image is rotated or the size is different, there is a problem that the correlation output sharply decreases and optical pattern recognition cannot be performed substantially.

In order to solve such a drawback, a hologram created by multiplexing reference images having various rotation directions and sizes at once is used as an optical filter, or rotation or size change is caused by a computer. Efforts have been made to create a composite discriminant function, which is a computer-generated hologram that reduces the decrease in correlation output even if it is used, and to print this on a photographic plate or the like as an optical filter. However, in these methods, a correlation output can be obtained only for a specific value with respect to rotation or change in size, and at the same time, creating an optical filter is complicated, so real-time light pattern recognition is It was totally unsuitable.

Further, even if the correlated image is rotated or its size is changed, the image is optically subjected to coordinate conversion into an image which does not substantially rotate or change its size, and the coordinate-converted image is converted. Attempts have been made to perform an optical Fourier transform correlation method using a correlation image and a reference image. As such a coordinate conversion method, a polar coordinate conversion method, an Lnr-θ conversion method, a Hough conversion method, and the like are known. These coordinate conversion methods are essentially based on the polar coordinate conversion method, and have the problem that the output periodically changes greatly with the change of the rotation angle due to the multi-value of the angle component. In order to perform the above, the computer-generated hologram described above must be created. Also,
Since this coordinate transformation must be performed as the first stage processing of the optical Fourier transform correlator, there is also a problem that a computer-generated hologram having extremely high diffraction efficiency must be created.

An object of the present invention is to accurately compare an input correlated image with the reference image even if the image to be correlated is rotated or has a different size. It is possible to provide an optical pattern recognition method and an optical pattern recognition device that can perform matching in real time.

[0008]

In order to solve the above-mentioned problems, an optical pattern recognition method or an optical pattern recognition apparatus according to the present invention uses a conventional Fourier transform correlator optical filter to add a Fourier transform hologram of a reference image. Alternatively, the low frequency component and the high frequency component of the joint Fourier transform hologram of the reference image and the correlated image in which the low frequency component and the high frequency component are individually recorded are used, and the optical filter in which these low frequency components are recorded and the high frequency component are recorded. A correlation output is obtained independently using an optical filter, an intensity ratio between the obtained correlation output from the low-frequency component and the correlation output from the high-frequency component is calculated, and a reference image or two reference images are used in advance. The optical pattern is obtained by comparing the intensity ratio between the obtained correlation output from the low-frequency component and the correlation output from the high-frequency component. It was identified. Since the intensity ratio of the correlation output has a unique value with respect to the image and has a property that it is hard to change with respect to rotation or a change in the size of the correlated image, the optical pattern recognition method and the optical pattern recognition method of the present invention The optical pattern recognition device can perform pattern recognition that is strong against rotation and size change of the correlated image with a simple method and configuration, and has solved the above-mentioned problem.

Further, real-time optical pattern recognition is realized by using an optical writing type ferroelectric liquid crystal spatial light modulator or a liquid crystal television as an optical filter.

[0010]

The transmission amplitude function of the correlated image is f, the transmission amplitude function of the reference image is g, and the transmission amplitude functions of these Fourier transform images are F (k) and G (k), respectively. At this time, the correlated function f rotates or the magnitude changes, and F
It is assumed that (k) is transformed to F (k + Δν). Here, Δν is a general deformation parameter in wavenumber space. At this time, when Δν takes a sufficiently small value, the transmission amplitude function F (k + Δν) of the deformed correlated function can be expressed as follows using the transmission amplitude function F (k) before deformation.

F (k + Δν) = F (k) + (∂F / ∂k) (dk / dν) Δν (1) Accordingly, an optical filter composed of a hologram created by the Fourier transform G (k) of the reference image in the Fourier transform after filtering (matched filter optical correlation processing), the output Pm _{is, P m = f ☆ g +} Δν · δ m (f ☆ g) + f * g + Δν · δ m (f * g) + DC (2) Becomes

Here, * is a correlation operator, * is a convolution operator, and DC is a DC component. Δ _m (f ☆ g) = ∫ (∂ / ∂ν) | F ^* (x−k) G (k) | ² · exp (−ikx) dk (3) δ _m (f * g) = ∫ (∂ / ∂ν) | F ^* (k−x) G (k) | ² · exp (−ikx) dk (4)

Further, after the transformed correlated image and the reference image are simultaneously Fourier-transformed to obtain a joint Fourier-transformed intensity distribution image, and then Fourier-transformed again (joint-transform correlation processing), the output P is P _j = f ☆ g + (Δν / 2) · δ _j (f ☆ g) + DC (5)

Here, δ _j (f ☆ g) = ∫ (∂ / ∂ν) [| F ^* (x−k) G (k) | ² + | G ^* (x−k) | ² ] exp ( −ikx) dk (6)

Here, the first and second terms of the equations (2) and (5), that is, the correlation output section are composed of a low frequency component and a high frequency component. That is, f ☆ g + Δν · δ _m (f ☆ g) = (f ☆ g) _L + Δν · δ _m (f ☆ g) _L + (f ☆ g) _H + Δν · δ _m (f ☆ g) _H (7) f ☆ g + (Δν / 2) · δ _j (f ☆ g) = (f ☆ g) _L + (Δν / 2) · δ _j (f ☆ g) _L + (f ☆ g) _H + (Δν / 2) ) · Δ _j (f ☆ g) _H (8)

Therefore, the ratio of the low-frequency component to the high-frequency component of the correlation output is calculated as follows: ρ _m = [(f ☆ g) _H + Δν · δ for each of the matched filter type optical correlator and the joint transform correlator. _m (f ☆ g) _H ] / [(f ☆ g) _L + Δν · δ _m (f ☆ g) _L ] (9) ρ _j = [(f ☆ g) _H + (Δν / 2) · δ _j ( f * g) _H ] / [(f * g) _L + ([Delta] v / 2) * [delta] _j (f * g) _L ] (10)

Here, ρ _m is the ratio between the low frequency component and the high frequency component of the correlation output in the matched filter type optical correlator, and ρ _j is the low frequency component and the high frequency component of the correlation output in the joint transform correlator. And the ratio. Therefore, when the deformation of the correlated image, that is, the deformation (∂F / ∂k) (dk / dν) Δν of the Fourier transform image is sufficiently small, ρ _m ρ _j （(fｆg) _H / (f ☆ g) ) _L = constant (11), and the ratio between the low frequency component and the high frequency component of the correlation output is an invariant. In other words, in the matched filter type correlator and the joint transform correlator, even if the correlated image is rotated or the size is changed, if these changes are sufficiently small, the low frequency component of the correlation output is obtained. And the ratio between the high frequency components is invariant.

Further, since the low-frequency component of the correlation output corresponds to the correlation with the rough features of the image, and the high-frequency component of the correlation output corresponds to the correlation with the delicate features of the image, these ratios are characteristic of the image itself. It is thought that it attaches. Therefore, the ratio between the low-frequency component and the high-frequency component of the correlation output obtained from the correlated image and the reference image is calculated by comparing the ratio between the low-frequency component and the high-frequency component of the correlation output obtained between the reference images in advance. , It is possible to perform pattern recognition that is strong against rotation and size change of the correlated image.

Next, a method of recording and modulating the low-frequency component and the high-frequency component of a Fourier transform hologram or a joint Fourier transform hologram independently using an optical writing type ferroelectric liquid crystal spatial light modulator will be described. I do. FIG. 5 is a schematic sectional view showing the structure of one embodiment of the optical writing type ferroelectric liquid crystal spatial light modulator used in the present invention.
In the optical writing type ferroelectric liquid crystal spatial light modulator shown in FIG. 5, a transparent electrode 55a such as ITO is formed on a transparent substrate 54a typically represented by quartz glass or the like having a flatness of λ / 8 or more. And furthermore, an intrinsic hydrogenated amorphous silicon (a-Si:
H) 56 are formed.

On the a-Si: H56, a dielectric mirror 57 is laminated in order to increase the gain of the reflectance of the reading light. The dielectric mirror 57 includes a λ / 4 film of silicon, a compound of silicon and germanium, or a high-refractive-index oxide such as titanium oxide or zirconium oxide, and a λ / 4 film of a low-refractive-index oxide such as silicon dioxide or magnesium fluoride.
/ 4 films are alternately stacked. On the other hand, on a transparent substrate 54b similar to the transparent substrate 54a, a transparent electrode 5 is provided.
A transparent electrode 55b similar to 5a is formed. On the transparent electrode 55b and the dielectric mirror 57, alignment films 54a and 54b made of SiO by oblique deposition or polyimide by rubbing are formed. These transparent substrates 54a and 54b are interposed by a ferroelectric liquid crystal 59 via a spacer 60 such that the alignment films 58a and 58b face the inner surface.
Is holding. In addition, anti-reflection coatings 61a and 61b are formed on the outer surfaces of the transparent substrates 61a and 61b.

The thus configured optical writing type ferroelectric liquid crystal spatial light modulator irradiates writing light 62 from the side of the transparent substrate 54a on which a-Si: H56 is formed to record an image.
The recorded image is read by irradiating the reading light 63 from the opposite transparent substrate 54b side. A continuous pulse voltage composed of a set of a positive voltage, a negative voltage, a positive voltage, and a zero voltage as shown in FIG. 7 is applied between the transparent electrodes 55a and 55b of the optical writing type ferroelectric liquid crystal spatial light modulator. Apply.

The first positive voltage has a pulse height Vp1 and a pulse width Tp1, the negative voltage has a pulse height Vn and a pulse width Tn, and the second positive voltage has a pulse height Vp2 and a pulse height Vp2.
It has a pulse width Tp2, and the zero voltage has a width Tz. At this time, the cycle of the continuous pulse voltage is Tp1 + Tn + Tp2 + Tz. Therefore, Vp1
When a negative voltage and a second positive voltage are applied so that> = Vn> Vp2, a threshold value for writing light (critical writing light intensity at which ferroelectric liquid crystal molecules are inverted and shifted to another stable state) Is
Since it is inversely proportional to the applied voltage, the threshold value Ithn of the writing light for the negative voltage is smaller than the threshold value Ithp2 for the second positive voltage, and the threshold value Ithp1 for the first positive voltage is smaller than the threshold value Ithn for the negative voltage. Or take an equal value.

First, the writing light intensity is larger than Ithn,
Let us consider the case where it is smaller than Ithp2. After the liquid crystal molecules are initialized in the one-way stable state at the first positive voltage, the liquid crystal molecules are inverted only at the portion irradiated with the writing light at the negative voltage, and an image is written. However, even when the second positive voltage is applied, the writing light having an intensity lower than the threshold value of the writing light for the second positive voltage is applied. Therefore, the writing light is applied when the second positive voltage is applied. The molecule does not return to its original stable state. That is, in such a case, the liquid crystal molecules in the portion irradiated with the writing light are in a stable state opposite to the stable state aligned in the initialization, so that the optical writing type ferroelectric liquid crystal spatial light modulation is performed. Images can be recorded on the container.

Next, let us consider a case where the writing light intensity is larger than Ithp2. At this time, as described above, the liquid crystal molecules are initialized in a one-way stable state at the first positive voltage, and then the liquid crystal molecules are inverted only at a portion irradiated with the writing light at the negative voltage, and an image is written. However, when the second positive voltage is applied,
Since the writing light having a light intensity higher than the writing light threshold is irradiated, the liquid crystal molecules in the portion irradiated with the writing light are inverted to the stable state aligned by initialization. That is, in such a case, since the liquid crystal molecules in the portion irradiated with the writing light have the same stable state as the stable state aligned by initialization, the optical writing type ferroelectric liquid crystal spatial light modulator is used. Can not record any images.

When the writing light intensity is weaker than Ithp1,
It goes without saying that no image can be recorded on the optical writing type ferroelectric liquid crystal spatial light modulator. Thus, FIG.
By continuously applying a pulse voltage as shown in FIG. 6, the optical writing type ferroelectric liquid crystal spatial light modulator has a threshold characteristic with respect to the writing light intensity as shown in FIG. In other words, when the writing light intensity is gradually increased while the same drive pulse voltage is continuously applied, the optical writing type ferroelectric liquid crystal spatial light modulator is turned on at a certain writing light intensity level. Reflected light intensity sharply increases (an image can be recorded).

As the writing light intensity is further increased, the reflected light intensity from the optical writing type ferroelectric liquid crystal spatial light modulator rapidly decreases at a certain writing light intensity level (an image is erased). ). Thereafter, when the writing light intensity is decreased, the intensity of the reflected light from the optical writing type ferroelectric liquid crystal spatial light modulator sharply increases again at a certain writing light intensity level (an image can be recorded again. Will be).

When the writing light intensity is further reduced, the intensity of the reflected light from the optical writing type ferroelectric liquid crystal spatial light modulator is reduced again from a certain writing light intensity level (the image is erased). ). Since ferroelectric liquid crystals have a distinct bistability, the reflected light intensity ideally changes to zero or one, but in practice, the alignment non-uniformity and the writing light non-uniformity are non-uniform. Therefore, the curve indicating the change in the reflected light intensity is slightly inclined at the change portion, and the threshold level cannot be specified. Therefore, the writing light intensity at which the reflected light intensity becomes 0.1 and 0.9 is defined as a threshold value of the optical writing type ferroelectric liquid crystal spatial light modulator with reference to the point where the reflected light intensity becomes maximum.

In the general example shown in FIG. 6, the threshold is IL1,
It takes a total of six values, IL2, IM1, IM2, IH1, and IH2,
Since IL1 to IL2, IM1 to IM2, and IH1 to IH2, IL
= IL1 to IL2, IM = IM1 to IM2, IH = IH1 to IH2. In the present invention, IL is a first threshold,
IH will be used as the second threshold value.

The threshold characteristic of the optical writing type ferroelectric liquid crystal spatial light modulator shown in FIG. 6 with respect to the writing light has a hysteresis and its cause is unknown, but does not affect the present invention. Furthermore, not only the optical writing type ferroelectric liquid crystal spatial light modulator but also a structure as shown in FIG. 5 using a light modulation material having bistability against an applied voltage instead of the ferroelectric liquid crystal. Needless to say, the same threshold characteristics for writing light as described above can be obtained even when the optical writing type spatial light modulator is used.

Now, when a Fourier transform image is modulated while a continuous driving pulse voltage as shown in FIG. 7 is applied to the optical writing type ferroelectric liquid crystal spatial light modulator as shown in FIG. FIG. 8 shows the intensity distribution of the Fourier transform image written in the optical writing type ferroelectric liquid crystal spatial light modulator.

FIG. 8A shows a Fourier-transformed image intensity distribution as a writing image, and shows a case where the visibility is low. In such a case, if the writing light intensity is sufficiently strong, the light intensity of the low frequency component of the Fourier transform image will be greater than the second threshold value IH. Therefore, at this time, the intensity distribution of the image written and read out to the optical writing type ferroelectric liquid crystal spatial light modulator is shown in FIG.
And only high frequency components are read out. The Fourier-transformed image with low visibility as shown in FIG. 8A is easily obtained by Fourier-transforming an image with a reduced contrast ratio or by Fourier-transforming the image using laser light with low coherence. Can be

Further, a Fourier transform image having high visibility as shown in FIG. 8B is applied to an optical writing type ferroelectric liquid crystal spatial light modulator while applying a continuous driving pulse voltage as shown in FIG. In writing, if the writing light is sufficiently strong, a read image as shown in FIG. 8D is obtained. That is, the high frequency component of the Fourier transform image is faithfully binarized and written in the optical writing type ferroelectric liquid crystal spatial light modulator, but only the edge portion of the low frequency component of the Fourier transform image is written. Will be. Such a Fourier-transformed image with high visibility can be easily obtained by Fourier-transforming an image with a high contrast ratio with laser light having high coherence. As described above, only the high frequency components of the Fourier transform image can be selected and written to the optical writing type spatial light modulator.

Further, the Fourier-transformed image is formed on the optical-writing type ferroelectric liquid crystal spatial light modulator at a writing light intensity such that the maximum value of the light intensity distribution of the Fourier-transformed image becomes smaller than the second threshold value IH. By writing, only the low frequency components of the Fourier transform image can be written. At this time, generally,
The high frequency component of the Fourier transform image is 1/1 of the low frequency component
This is because, since it is weak to about 0 to 1/1000, it becomes smaller than the first threshold value IL of the optical writing type ferroelectric liquid crystal spatial light modulator, and writing cannot be performed. As described above, only the high frequency components of the Fourier transform image can be selected and written to the optical writing type spatial light modulator.

Of course, as a method of selectively recording the low-frequency component and the high-frequency component of the Fourier transformation described above in the optical writing type ferroelectric liquid crystal spatial light modulator, the writing light intensity is not changed. It goes without saying that the above may be performed by changing the pulse height of the second positive voltage in FIG.

[0035]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of an optical pattern recognition method and an optical pattern recognition apparatus according to the present invention will be described below with reference to the drawings. FIG. 1 is a process diagram showing an embodiment in which the optical pattern recognition method of the present invention is configured based on a joint transform correlator. In FIG. 1, a correlated image 3 captured from an imaging device 1 such as a CCD camera or a vidicon.
And the RAM (Random Access) in the computer
ss Memory) or ROM (Read Only)
Reference image read out from a memory such as a Memory, an external memory (a hard disk, a floppy disk, or an optical disk) or a memory such as a photographic film is simultaneously Fourier-transformed by coherent light (joint Fourier transform 5) to form a joint Fourier transformed image. . Here, the joint Fourier transform image is an interference fringe distribution formed by overlapping the Fourier transform image of the correlated image and the Fourier transform image of the reference image on the optical axis and interfering with each other, and is a kind of Fourier transform hologram. It is.

The joint Fourier transform image obtained in this manner is subjected to a low frequency component 6 and a high frequency component 11 by using, for example, an optical writing type ferroelectric liquid crystal spatial light modulator as described in the operation. Are independently obtained from the intensity distribution images 7 and 12
Convert to The intensity distribution image 7 of the low-frequency component 6 of the joint Fourier transform 5 and the intensity distribution image 12 of the high-frequency component 11 of the joint Fourier transform image 5 are Fourier-transformed again independently (8, 13) to become correlation images. The correlation output of the correlation image is detected by a CCD camera, a vidicon, or a photodiode (9, 14). A correlation coefficient is calculated by a computer or an electronic logic using the correlation peak intensity measured from the detected correlation output (10, 15). The series of steps 6 to 15 for obtaining the correlation coefficient may be sequentially performed in chronological order, or the steps 6 to 10 and the steps 11 to 15 may be substantially performed in parallel with two joint transform correlators. The operations may be performed in parallel by operating them in parallel. Using the correlation coefficient corresponding to the low-frequency component 6 of the joint Fourier transform 5 and the correlation coefficient corresponding to the high-frequency component 11 of the joint Fourier transform 5 obtained in this manner, the correlation is performed by computer or electronic logic. A relation number ratio is calculated (16).

On the other hand, the memory 17 obtains the joint Fourier transform in the above-described process using the same two reference images, and obtains the correlation coefficient obtained from the low-frequency component and the correlation coefficient obtained from the high-frequency component. The ratio with the correlation coefficient is
It is stored in association with the reference image. The correlation coefficient ratio 18 read from the memory 17 is compared with the correlation coefficient ratio obtained in the above step 16 and compared (19) to determine whether the difference or ratio of these numerical values is within a predetermined value. Thus, it is determined whether the correlated image 3 and the reference image 4 are the same (20). The numerical range serving as the determination criterion is comprehensively determined from the SN ratio of the detected correlation output, the type of the reference image (the difference between images such as characters and landscapes), the number of reference images, the determination probability, and the like.

FIG. 3 shows the optical pattern recognition method of the present invention shown in FIG. 1 more specifically. FIG. 3 is a configuration diagram showing the configuration of an embodiment of the optical pattern recognition device of the present invention, wherein 28 is a writing laser, 29 is a first beam expander, 30 is a first CCD camera, 31 and 44 are A / D converters, 32 is a computer, 33 is a D / A converter, 34 is a liquid crystal television driving circuit, 35 is a liquid crystal television, 36 is a first Fourier transform lens, and 37 is an optical writing type strong. A dielectric liquid crystal spatial light modulator, 38 is a reading laser,
39 is a second beam expander, 40 is a polarization beam splitter, 41 is a second Fourier transform lens, 42 is a second CCD camera, and 43 is a video monitor.

In FIG. 3, a correlated image including a target image input from a first CCd camera 30 is converted into a digital signal by an A / D converter 31 and then taken into a computer 32. In the computer 32, the reference image previously stored in the memory therein and the fetched image to be correlated are synthesized adjacent to each other, and this is again converted into an analog signal by the D / A converter 33, and The image is displayed on the liquid crystal television 35 via the drive circuit 34.

FIG. 11 shows an example of an image displayed on the liquid crystal television 35 in this manner. FIG. 11 shows an example in which alphabetic characters B are displayed adjacently on a liquid crystal television as a correlated image and a reference image. As shown in FIG. 11, generally, a correlated image is input after being rotated or different in size from a reference image. In this example, the reference image is rotated by 10 degrees with respect to the correlated image, and is reduced by about 36%. Also, in FIG. 1, two liquid crystal televisions are arranged adjacent to each other instead of the liquid crystal television 35, and the correlated image from the first CCD camera 30 is placed on one of the liquid crystal televisions, and the computer 3 is placed on the other liquid crystal television.
Needless to say, the reference image from No. 2 may be directly displayed.

In FIG. 1, the coherent light emitted from the writing laser 28 enters the first beam expander 2.
After the beam diameter is converted to an appropriate size by 9, the correlated image and the reference image displayed on the liquid crystal television 35 are simultaneously read and converted to a coherent image. In this embodiment, the writing laser 28 has a wavelength of 6 which can be easily directly modulated.
The correlated image and the reference image converted into a coherent image using a semiconductor laser of 90 nm are simultaneously subjected to Fourier transform sale and joint Fourier transform by a first Fourier transform lens 36 as an optical writing type ferroelectric liquid crystal spatial light. The writing surface of the modulator 37 is irradiated. At this time, a driving pulse voltage as shown in FIG. 7 is applied to the optical writing type ferroelectric liquid crystal spatial light modulator 37. As the driving pulse voltage, the first positive voltage has a pulse height of 10 V and a pulse width of 2
msec, negative voltage having a pulse height of 10 V and pulse width of 1 msec, second positive voltage having a pulse height of 7 V and a pulse width of 0.8 msec, and a period of 33
msec.

On the other hand, the coherent light emitted from the reading laser 38 is expanded to an appropriate beam diameter by the second beam expander 39 and then reflected by the polarization beam splitter 40 to be written in the optical writing type ferroelectric liquid crystal space. The joint Fourier transform image written in the optical modulator 37 is read. As the reading laser 38, a semiconductor laser having the same wavelength of 690 nm as the writing laser was used. The read joint Fourier transform image passes through the polarizing beam splitter 40 again and is converted into an intensity distribution image, and is then Fourier transformed again by the second Fourier transform lens 41 to become a correlation image and becomes a second CCD camera 42 Is imaged and converted into an electric signal. The correlation image converted into an electric signal is directly observed on a video monitor 43 or converted into a digital signal by an A / D converter 44 and taken into the computer 32. The maximum value of the correlation peak included in the correlation image captured by the computer 32 is detected and stored in a memory in the computer 32 as a correlation coefficient.

The writing light is irradiated during a period in which the first positive voltage, the negative voltage, and the second positive voltage are applied in the driving pulse voltage of the optical writing type spatial light modulator 37 shown in FIG. The reading light was applied during the period of zero voltage. By doing so, it is possible to prevent the reading light from affecting the writing of the joint Fourier transform image. Of course, when the reading light is weak enough not to affect the writing of the joint Fourier transform image, it is not necessary to apply such modulation to both the writing light and the reading light. When a laser light source such as a semiconductor laser, which is difficult to directly modulate, is used, the writing laser 28 and the optical writing type ferroelectric liquid crystal spatial light modulator 3 are used.
7, an external modulator such as a liquid crystal shutter is set between the reading laser 38 and the optical writing type ferroelectric liquid crystal spatial light modulator 37 to modulate the writing light and the reading light, respectively. Needless to say, this may be done.

The liquid crystal television 35 and the optical writing type ferroelectric liquid crystal spatial light modulator 37 are positioned at the front focal plane and the rear focal plane of the first Fourier transform lens 36, respectively. It is needless to say that the liquid crystal spatial light modulator 37 and the second CCD camera 42 are arranged at the front focal plane and the rear focal plane of the second Fourier transform lens 41, respectively.

In the joint conversion correlator shown in FIG. 1, first, in order to obtain a correlation coefficient corresponding to a low frequency component, the writing light intensity is changed to the optical writing type ferroelectric liquid crystal spatial light modulator 3.
Immediately before Step 7, the output of the writing laser 28 was adjusted so that the average was about 290 μW / cm ² . The read light intensity was set to 270 μW / cm ² . FIG. 12A shows an example of the low-frequency component of the joint Fourier transform image written in the optical writing type ferroelectric liquid crystal spatial light modulator 37 in this manner. FIG. 12A shows an example of a low-frequency component of a joint Fourier transform image obtained by adjoining two alphabetic characters B having the same orientation and size.
Although an interference fringe corresponding to a carrier wave is superimposed on the joint Fourier transform image, it is omitted in FIG. The correlation coefficient obtained by Fourier-transforming the joint Fourier-transformed image thus obtained is stored in the memory of the computer 32.

Next, in order to obtain the correlation coefficient corresponding to the high frequency component, the writing light intensity is about 2.2 mW / cm 2 on average just before the optical writing type ferroelectric liquid crystal spatial light modulator 37. The output of the writing laser 28 was adjusted as described above. FIG. 12B shows an example of the joint Fourier transform image written in the optical writing type ferroelectric liquid crystal spatial light modulator 37 in this manner. Similarly to FIG. 12A, one of the high-frequency components of the joint Fourier transform image obtained by adjoining two alphabetic characters B having the same orientation and size to each other.
It is an example. The correlation coefficient obtained by Fourier-transforming the joint Fourier-transformed image thus obtained is stored in the memory of the computer 32.

The correlation coefficient corresponding to the low frequency component and the correlation coefficient corresponding to the high frequency component of the correlated image and the reference image stored in the memory of the computer 32 are read out in the computer 32. Then, a correlation coefficient ratio, which is a ratio of these numerical values, is calculated. The correlation coefficient ratio obtained when two alphabetic characters B having the same orientation and size were adjacent to each other was about 1.55. When the value of the correlation coefficient corresponding to the low frequency component is plotted on the graph against the value of the correlation coefficient corresponding to the high frequency component for the uppercase alphabetic characters other than B, the plot for the following character set is plotted. However, it turned out to be on a straight line having almost the same inclination. That is, (F, A,
R), (V, K, H, N, D), (P, Y, I),
(T, L, J), (X, S, E, B), (O, C, Q,
U, G) and (M, Z, W). It can be seen that these characters are composed of relatively similar component elements (similar characters).

This means that the optical pattern recognition method or optical pattern recognition apparatus of the present invention can relatively well separate and process low frequency components and high frequency components of a correlated image and a reference image. It is considered to indicate.
By the way, the inclination of the straight line on which these character sets are riding is
It was about 0.8. It should be noted that the inclination of the straight line slightly changes depending on the conditions of the driving pulse voltage of the optical writing type ferroelectric liquid crystal spatial light modulator 37 and the writing light intensity.

Next, it will be described that the optical pattern recognition method and the optical pattern recognition apparatus of the present invention are a method and a system which are resistant to rotation and a change in size of a correlated image. To explain this, an example of an embodiment performed using the optical pattern recognition device of the present invention based on the joint transform correlator shown in FIG. 3 will be described with reference to the drawings. Also in the present embodiment, uppercase alphabetic characters are used for the correlated image and the reference image.

FIG. 13 shows the correlation coefficient 75 from the low frequency component, the correlation coefficient 76 from the high frequency component, and the correlation coefficient ratio 74 when the correlated image is rotated without changing its size.
6 is a graph showing how the value changes. FIG.
Uses a capital letter B as a correlated image and a reference image.

The values of the correlation coefficient 75 from the low-frequency component and the correlation coefficient 76 from the high-frequency component in FIG. 13 are standardized based on the value when the correlated image is not rotated. . As is clear from this graph, when the correlated image rotates, the correlation coefficient 75 corresponding to the low-frequency component and the correlation coefficient 76 corresponding to the high-frequency component decrease sharply,
When the correlated image is rotated by about 12 degrees, these values become about half. On the other hand, the value of the correlation coefficient ratio 74 increases only by about 10% even when the correlated image is rotated by about 20 degrees. In other words, it can be seen that the use of the correlation coefficient ratio 74 enables pattern recognition extremely resistant to rotation of the correlated image.

The correlation coefficient ratio 74 has a value peculiar to the image, and the pattern can be recognized using the value. For example, even if there are two images having a correlation coefficient ratio with respect to the autocorrelation which are extremely close to each other, once the reference image is specified, these can be easily determined. This is because the correlation coefficient ratio for the cross-correlation of these images takes a completely different value from the correlation coefficient ratio for the autocorrelation. Two images having exactly the same correlation coefficient ratio for autocorrelation and correlation coefficient ratio for cross-correlation are the same image.

Next, a case where the size of the correlated image is different from the size of the reference image will be described. FIG.
Reference numeral 4 denotes a correlation coefficient 75 corresponding to a low-frequency component when the size of a correlated image changes by the optical pattern recognition device of the present invention based on the joint transform correlator shown in FIG. Correlation coefficient 7 corresponding to frequency component
6 is a graph showing changes in the correlation coefficient ratio 74. In FIG. 14, as in FIG. 13, a capital letter B is used as the correlated image and the reference image.
Further, the values of the correlation coefficient 75 from the low-frequency component and the correlation coefficient 76 from the high-frequency component are standardized on the basis of the value when the size of the correlated image does not change.

As the amount of change in the size of the image, the absolute value of the scale change rate is plotted on the horizontal axis of the graph. That is, scale change rate = (area of image whose size has changed−area of image whose size has not changed) ×
100 / (area of the image whose size has not changed).

As is apparent from FIG. 14, in this embodiment, when the size of the correlated image changes, the correlation coefficient ratio 74 also changes, and when the scale changes by about 10%, the correlation coefficient ratio 74 also changes by about 10%. You can see that This means that the scale is about 5
When it changes by ７7%, the correlation coefficient 75 corresponding to the low frequency component
And the correlation coefficient 76 corresponding to the high frequency component is about 10%
It can be seen from the change that the pattern recognition ability is improved as compared with the conventional optical pattern recognition device or optical pattern recognition method, but it is not so much improvement in characteristics. This is because the visibility of the joint Fourier transform between the correlated image and the reference image suddenly deteriorates due to the change in the size of the correlated image, and the joint Fourier transform image is converted to a light-writing type ferroelectric liquid crystal spatial light modulation. This is because the joint Fourier transform image cannot be sufficiently written in the optical writing type ferroelectric liquid crystal spatial light modulator due to the binarization by the light modulator.

The decrease in the visibility of the joint Fourier transform image is caused by a change in the size of the correlated image, resulting in a poor balance of the optical power density between the Fourier transform image of the correlated image and the Fourier transform image of the reference image. It is caused by things. In fact, by applying a positive pulse voltage and a negative pulse voltage of sufficiently high frequency alternately to the optical writing type ferroelectric liquid crystal spatial light modulator, an image having continuous gradation can be obtained. It is known that it is possible to display an image on a dielectric liquid crystal spatial light modulator. By using this driving method, a joint Fourier transform image of a correlated image having a change in size and a reference image can be displayed on an optical writing type. When a graph similar to that of FIG. 14 was obtained by displaying on the dielectric liquid crystal spatial light modulator,
It has been found that the correlation coefficient ratio changes only about 10% even if the size of the correlated image changes by 20 to 24%. That is, essentially, the optical pattern recognition method or optical pattern recognition device of the present invention can accurately perform pattern recognition even if the correlated image or the reference image rotates or changes in size.

However, by binarizing the joint Fourier transform image, a correlation output with a high SN ratio can be obtained, and the driving method of the optically-written ferroelectric liquid crystal spatial light modulator does not cause a problem. Therefore, even if the size of the correlated image changes, the gradation is set so that the light intensity transmitted through the correlated image and the reference image displayed on the liquid crystal television shown in FIG. The correlated image and the reference image were displayed on the liquid crystal television 35, and a graph similar to that of FIG. 14 was created. Such display of the reference image and the correlated image can be achieved by changing the brightness of the portion of the liquid crystal television 35 where the image is displayed in inverse proportion to the size of the image.

In this embodiment, since the sizes of the correlated image and the reference image are known in advance, the computer 32
The output was directly changed. Practically, immediately after the means for inputting the correlated image and the reference image such as the liquid crystal television 35 to the optical system, means for partially splitting light such as a beam splitter is provided, and the light of the split light is provided. Of these, the size of each image is measured by independently collecting and measuring the light corresponding to the correlated image and the reference image,
It is practical to determine the brightness between the correlated image displayed on the liquid crystal television 35 and the reference image. At this time, it is important that the background of the image to be displayed (other than the image) does not transmit light as much as possible so that the visibility of the joint Fourier transform image is not reduced.
In this way, keeping the input intensity of the correlated image and the reference image the same and suppressing a decrease in the visibility of the joint Fourier transform image is referred to as input image normalization.

FIG. 15 shows the phase corresponding to the low-frequency component with respect to the scale change rate of the correlated image when the above-described input image is normalized and the optical pattern recognition device shown in FIG. 3 is operated. The relation between the relation number 75 and the correlation coefficient 76 and the correlation coefficient ratio 74 corresponding to the high frequency component is shown in a graph. As is clear from FIG. 15, the correlation coefficient 75 corresponding to the low frequency component and the correlation coefficient 76 corresponding to the high frequency component decrease to about half when the scale change rate becomes about 12%. The correlation coefficient ratio 74 shows a change of 10% only when the scale change rate becomes about 22%.

This is slightly inferior in performance to the case where the joint Fourier transform image is written as a continuous tone image in the optical writing type ferroelectric liquid crystal spatial light modulator, but is more inferior to the conventional joint transform correlator. It can be seen that the optical pattern recognition device of the present invention has a very good pattern recognition capability with respect to a change in the size of a very correlated image or a reference image. That is, by standardizing and inputting the correlated image or the reference image, the optical pattern recognition device or the optical pattern recognition method of the present invention based on the binarized joint transform correlator can perform Even if the size of the image changes, it exhibits excellent pattern recognition ability.

The optical pattern recognition method and optical pattern recognition apparatus of the present invention have been described above using an optical system based on a joint transform correlator. Hereinafter, an embodiment of an optical pattern recognition method and an optical pattern recognition device of the present invention will be described using an optical system based on a matched filter type optical correlator. FIG.
FIG. 3 is a process diagram showing one embodiment of the optical pattern recognition method of the present invention based on a matched filter type optical correlator.

In FIG. 2, parts having the same functions as those in FIG. 1 are denoted by the same reference numerals. In FIG. 2, the reference image 4 read from the memory 2 is converted into a coherent image by coherent light such as laser light, and is optically Fourier-transformed (22). The optically Fourier transformed reference image is superimposed on a coherent carrier wave (usually a coherent plane wave) and recorded as a Fourier transform hologram on a photographic film or an optical writing type spatial light modulator. At this time, the recorded Fourier transform hologram is formed only from the high frequency component or the low frequency component (24, 25). The recording of only the low-frequency component or the high-frequency component of the Fourier transform hologram can be performed, for example, by using the threshold characteristics of the optical writing type ferroelectric liquid crystal spatial light modulator as described in the operation.

On the other hand, a correlated image 3 read from an imaging device (may be a memory in a computer) 1 such as a CCD camera is converted into a coherent image by coherent light such as a laser, and then optically Fourier-transformed. (21). The Fourier transform of the correlated image is frequency-filtered independently by the low frequency component of the Fourier transform hologram of the reference image and the high frequency component of the Fourier transform hologram of the reference image (26,
27). The Fourier-transformed images of the correlated images, which have been frequency-filtered by the low-frequency component hologram and the high-frequency component hologram, are again independently Fourier-transformed and converted into correlation functions, and their correlation outputs are detected (9, 1).
4). The subsequent pattern recognition method is the same as the method based on the joint transform correlator described with reference to FIG. However, the correlation coefficient ratio stored in the memory 17 is a correlation coefficient ratio obtained by using a reference image instead of a correlated image in steps other than 17 to 20 in FIG.

The difference between this method and the method based on the joint transform correlator is that the visibility of interference fringes when forming a Fourier transform hologram has a high value only for a specific light intensity region. And that the selection of the carrier component is relatively flexible within the range of the resolution of the spatial light modulator. In addition, since the correlated image and the reference image are separately subjected to Fourier transform, it is not necessary to combine them at the time of input, or to use two electric writing type spatial light modulators such as a liquid crystal television in parallel. That is.

FIG. 4 shows an embodiment of the optical pattern recognition apparatus of the present invention based on the matched filter type optical correlator shown in FIG. In FIG. 4, 44 is an A / D converter, 45 is a first beam splitter, 46 is a first liquid crystal television, 47 is a second beam splitter, 48 is a first mirror, 49 is a second mirror, Reference numeral 50 denotes a second liquid crystal television, 51 denotes a third Fourier transform lens, 52 denotes a first liquid crystal television driving circuit, and 53 denotes a second liquid crystal television driving circuit.

In FIG. 4, components having the same functions as in FIG. 3 are denoted by the same reference numerals, and description thereof is omitted. In FIG. 4, the reference image stored in the memory of the computer 32 is converted into an analog electric signal by the D / A converter 33, and the first electric signal is transmitted through the first liquid crystal television driving device 52.
Is displayed on the liquid crystal television 46.

Next, the laser beam emitted from the writing laser 28 is expanded to an appropriate beam diameter by the first beam expander 29, and a part of the laser beam is transmitted through the first beam splitter. Then, the reference image displayed on the first liquid crystal television 46 is irradiated to convert it into a coherent light image. The reference image converted into the coherent light image is transmitted through the second beam splitter 47 to the first image.
The Fourier transform lens 36 performs Fourier transform on the writing surface of the optical writing type ferroelectric liquid crystal spatial light modulator 37.

On the other hand, the laser light partially reflected by the first beam splitter 45 is successively reflected by the first mirror 48, the second mirror 49, and the second beam splitter 47, and is written in an optical writing type. The writing surface of the ferroelectric liquid crystal spatial light modulator 37 is irradiated as a plane wave. The Fourier transform image of the reference image and the plane wave interfere with each other on the writing surface of the optical writing type ferroelectric liquid crystal spatial light modulator 37 to form interference fringes,
The Fourier transform hologram of the reference image is recorded on the optically-writable ferroelectric liquid crystal spatial light modulator 37. The plane wave corresponds to a carrier wave when forming the Fourier transform hologram, and the magnitude of the carrier wave component is uniquely determined by an angle between the optical axis of the plane wave and the optical axis of the Fourier transform hologram of the reference image. It is determined.

As in this embodiment, when the optical writing type ferroelectric liquid crystal spatial light modulator is used as a medium for recording the Fourier transform hologram of the reference image, the optical axis of the plane wave and the optical axis of the reference image are used. The angle between the Fourier transform hologram and the optical axis is about 9 degrees or less. At this time, similarly to the embodiment shown in FIG. 3, a drive voltage waveform as shown in FIG. 7 is applied to the optical writing type ferroelectric liquid crystal spatial light modulator 37, and the output of the writing laser 28 is appropriately adjusted. As a Fourier transform hologram of the reference image recorded in the optical writing type ferroelectric liquid crystal spatial light modulator 37, a Fourier transform hologram having the low frequency component as a main component, A Fourier transform hologram having a frequency component as a main component can be obtained independently of each other.

On the other hand, the correlated image including the target image is
After being imaged by the first CCD camera 30 and converted into an electric signal, it is displayed on the second liquid crystal television 50 via the second liquid crystal television driving circuit 53. Then, the laser beam emitted from the reading laser 38 is converted into an appropriate beam diameter by the second beam expander 39,
The correlated image displayed on the second liquid crystal television 50 is radiated and converted into a coherent optical image. The correlated image converted into the coherent optical image is converted by the third Fourier transform lens 51 through the polarization beam splitter 40 into the optically-written ferroelectric liquid crystal spatial light modulator 3.
7 is Fourier-transformed on the read-out surface 7. At this time, it is necessary to make the optical axis of the Fourier transform of the reference image coincide with the optical axis of the correlated image as much as possible in order to perform pattern recognition with a high SN ratio.

The optical image thus read out is diffracted in the same optical axis direction as the plane wave. This readout image does not deviate from the result of frequency filtering of the Fourier transform of the correlated image with the Fourier transform hologram of the reference image. The Fourier-transformed image of the correlated image subjected to the frequency filtering is converted into a light intensity image by transmitting through the polarization beam splitter 40, and is converted by the second Fourier transform lens 41 again into the photoelectric of the second CCD camera 42. Fourier transformation is performed on the transformation plane to become a correlation function and a convolution function between the correlated image and the reference image. The correlation function is photoelectrically converted and converted into a digital electric signal by the A / D converter 44, and then stored in the memory of the computer 32.

The correlation signal stored in the memory of the computer 32 is read out again in the computer 32, the maximum value of the correlation peak included in the correlation function is calculated and stored as a correlation coefficient in the memory of the computer 32 again. Is done. The correlation coefficient obtained at this time is a correlation coefficient corresponding to the low frequency component when the Fourier transform hologram of the reference image is mainly composed of low frequency components, and the Fourier transform hologram of the reference image is mainly When it is composed of high frequency components, it becomes a correlation coefficient corresponding to the high frequency components. Then, the correlation coefficient corresponding to the low-frequency component and the correlation coefficient corresponding to the high-frequency component are read from the memory of the computer 32, and the correlation coefficient ratio, which is the ratio, is calculated in the computer 32.

The correlation coefficient ratio between the correlated image and the reference image calculated in this manner is successively equal to the correlation coefficient ratio obtained from the reference images previously stored in the memory of the computer 32. By comparing and matching, it is possible to determine whether or not the correlated image and the reference image match.

As described above, in the same manner as shown in FIGS. 13 to 15, when the correlated image is rotated or the size is changed, Investigations were made to see if the characteristics were exhibited, and almost the same results were obtained. Of course, in FIG. 4, the input of the correlated image is performed by the first liquid crystal television 46, and the input of the reference image is performed by the second liquid crystal television 46.
It is clear that the same result can be obtained even when the liquid crystal television 50 is used.

Next, an embodiment in which an optical pattern recognition is performed using a CCD camera and a liquid crystal television as a recording medium of a Fourier transform hologram, instead of a light-writing type ferroelectric liquid crystal spatial light modulator, will be described. . FIG. 9 is a block diagram showing one embodiment showing this, 64 is a first laser, 65 is a third CCD camera, 66 is an A / D converter,
67 is a second laser, 68 is a third liquid crystal television, 69 is a D / A converter, and 70 is a third liquid crystal television drive circuit.

In FIG. 9, the components having the same functions as those in FIG. 4 are denoted by the same reference numerals, and the description thereof is omitted. In FIG. 9, an image including a correlated image captured from the first CCD camera 30 (hereinafter, referred to as a correlated image for simplicity) is first transmitted through a second liquid crystal television driving circuit 53 to a second liquid crystal display. It is displayed on the television 50. On the other hand, the coherent light emitted from the first laser 64 is converted into an appropriate beam diameter by the second beam expander 39,
And partially converts the correlated image displayed on the second liquid crystal television 50 into a coherent light image. The correlated image converted into the coherent optical image is converted into a second image by the third Fourier transform lens 51.
The third CCD camera 6 via the beam splitter 47
5 is Fourier-transformed on the photoelectric conversion surface 5. On the other hand, the coherent light partially reflected by the first beam splitter 45 is reflected by the first mirror 48, the second mirror 49, and the second beam splitter 47 one after another, and then becomes third light.
Is irradiated as a carrier wave on the photoelectric conversion surface of the CCD camera 65.

That is, similarly to the embodiment shown in FIG. 4, the Fourier transform of the correlated image and the carrier interfere with each other, and the Fourier transform of the correlated image is placed on the photoelectric conversion surface of the third CCD camera 65. The conversion hologram is formed. At this time, the angle formed by the optical axis of the Fourier transform of the correlated image and the optical axis of the carrier is determined by the third CCD camera 65.
Must be small enough to form an interference fringe that is sufficiently large compared to the pixel pitch.

The Fourier transform hologram of the correlated image picked up by the third CCD camera 65 in this manner is represented by A
The signal is converted into a digital electric signal by the / D converter 66 and stored in the memory of the computer 32. The computer 32 extracts only the high frequency region and the low frequency region of the Fourier transform hologram of the correlated image stored in the computer 32.

For example, a circular region having a specific radius from the optical axis is set, and the Fourier transform hologram of the correlated image inside the circular region is set to the low frequency of the Fourier transform hologram of the correlated image. And the remaining area may be the high frequency component of the Fourier transform hologram of the correlated image. The low-frequency component and the high-frequency component of the Fourier-transformed hologram of the extracted correlated image are independently converted into analog electric signals by the D / A converter 69, and then the third liquid crystal television drive The information is displayed on the third liquid crystal television 68 via the circuit 70. At this time, the low-frequency component and the high-frequency component of the correlated image displayed on the third liquid crystal television are different from an image having a continuous tone within a gamma characteristic range determined by the CCD camera, the liquid crystal television, or the electric circuit. Has become.

Further, the digital electric signal corresponding to the reference image is sent from the computer 32 to the D / A converter 33 and converted into an analog electric signal. 1 is sent to the liquid crystal television 46, and the reference image is displayed thereon.

The coherent light emitted from the first laser 67 is expanded to an appropriate beam diameter by the first beam expander 29, and then the reference image displayed on the first liquid crystal television 46 is displayed. Irradiate and convert this to a coherent light image. The reference image converted into the coherent optical image is Fourier-transformed by the first Fourier transform lens 36 and the low-frequency component of the correlated image or the high-frequency component of the correlated image displayed on the third liquid crystal television 68. Is irradiated. At this time, the first liquid crystal television 4
The sixth and third liquid crystal televisions 68 are disposed on the front focal plane and the rear focal plane of the first Fourier transform lens 36, respectively.
At this time, the relationship between the display surface of the third liquid crystal television 68 and the optical axis of the Fourier transform of the reference image is determined by the third CCD.
The most efficient frequency filtering can be performed when the relationship (the incident angle and the incident position) between the photoelectric conversion surface of the camera 65 and the optical axis of the Fourier transform of the correlated image is completely the same.

As described above, the Fourier transform of the reference image which has been frequency-filtered by the low-frequency component or the high-frequency component of the Fourier transform of the correlated image is Fourier-transformed again by the second Fourier transform lens 41, and the second CC
A correlation peak occurs on the photoelectric conversion surface of the D camera 42. Correlation peaks obtained independently from the low frequency component and the high frequency component of the Fourier transform of these correlated images are A /
The signal is converted into a digital electric signal by the D converter 44, and the correlation coefficient and the correlation coefficient ratio corresponding to each are obtained in the computer 32, as in the embodiment shown in FIG.

A description will be given of the result of pattern recognition using the alphabet characters as the correlated image and the reference image in the embodiment shown in FIG. At this time,
When the correlated image rotates about 6 degrees with respect to the reference image, the magnitude of the correlation coefficient corresponding to the low-frequency component decreases to about 、, but the value of the correlation coefficient ratio is It did not decrease by half until it rotated about 12 to 16 degrees with respect to the reference image.

Further, when the size of the correlated image changes by about 5% compared to the size of the reference image, the size of the correlation coefficient corresponding to the low frequency component decreases to about １ ／. The value of the correlation coefficient ratio is such that the correlated image is about 1 relative to the reference image.
It did not decrease by half until it changed by 2 to 15%. These results show that the characteristics are worse as compared with the case where an optical writing type ferroelectric liquid crystal spatial light modulator is used as a means for displaying a Fourier transform hologram or a joint Fourier transform image. The resolution of the LCD television used to display the hologram is approximately 9 l
This is because it has only a small value of ines pair / mm, and is not an essential problem. Rather, in the present embodiment, the effect that the low frequency component and the high frequency component of the Fourier transform hologram can be created with a large degree of freedom on software is important.

The embodiment in which the optical pattern recognition method and the optical pattern recognition apparatus of the present invention are applied to the optical correlation based on the Fourier transform has been described so far. An embodiment will be described which shows that the present invention can be applied to a method of performing optical pattern recognition by filtering an image with a Fourier transform image of a reference image and measuring the intensity thereof.

As shown in FIG. 10, the optical system for performing the optical pattern recognition is shown in FIGS.
A first beam splitter 45, a first mirror 48, a second
Only the mirror 49 and the second beam splitter 47 are not provided. Also, the second Fourier transform lens 41
Does not require the Fourier transform of the readout image, and uses the condenser lens 71. Therefore, what is captured by the third CCD camera 65 or recorded / displayed on the optically-writable ferroelectric liquid crystal spatial light modulator 37 is the low-frequency component of the Fourier transform hologram of the correlated image or the reference image. It is not a high frequency component but a low frequency component or a high frequency component of the Fourier transform of the correlated image or the reference image.

Further, what is imaged by the second CCD camera 42 and measured, calculated and collated in the computer 32 is not the correlation coefficient or the correlation coefficient ratio between the correlated image and the reference image, but It is the total intensity and the total intensity ratio of the low frequency component and the high frequency component of the Fourier transform image of the correlated image or the reference image subjected to the frequency filtering. That is, the optical writing type ferroelectric liquid crystal spatial light modulator 37 includes:
The negative image of the Fourier transform image of the reference image is recorded and displayed, and the Fourier transform image of the correlated image is reflected only from a part different from the Fourier transform pattern of the correlated image. The operation of the optical pattern recognition device according to the present embodiment is the same as that of the optical pattern recognition device according to the embodiment shown in FIGS.

As described above, the low-frequency component of the Fourier transform of the correlated image is converted from the optical writing type ferroelectric liquid crystal spatial light modulator 37 in which the high-frequency component of the Fourier transform image of the reference image is recorded. The high frequency component of the Fourier transform of the correlation image is obtained from the optical writing type ferroelectric liquid crystal spatial light modulator 37 in which the low frequency component of the Fourier transform of the reference image is recorded. Therefore, when the correlated image and the reference image completely match both in size and rotation direction, the total intensity obtained from the low-frequency component or the high-frequency component of the correlated image is minimized. The greater the difference between the correlated image and the reference image, the larger the difference between the correlated image and the reference image, and the larger the rotation angle of the reference image with respect to the correlated image. The larger, the stronger.

However, using the method of the present invention, the total intensity ratio, which is the ratio of the total intensity corresponding to the low-frequency component to the total intensity corresponding to the high-frequency component, is the size of the correlated image and the reference image. Even if they are different or they are in a mutually rotated relationship, if the differences and the rotation angle are sufficiently small values, they are kept almost constant. The difference in size is about 18 to 22%, and the rotation angle is about 1
From 8 to 20 degrees, the total intensity ratio was kept almost constant.
Furthermore, since the total intensity ratio indicates a value specific to the correlated image or the reference image used, optical pattern recognition can be performed by using this value.

The optical pattern recognition method shown in the present embodiment does not require an interference fringe as a carrier component on the Fourier transform surface unlike the matched filter type optical correlator and the joint transform correlator. Therefore, even if a low-resolution spatial light modulator such as a liquid crystal television is used as a medium for recording the Fourier transform of the correlated image or the reference image, the pattern recognition ability will not be significantly reduced.

In the above-described embodiment, an example in which alphabetic characters are used as the correlated image or the reference image has been described. However, other images may be used for the optical pattern recognition method or the optical pattern recognition apparatus according to the present invention. That can be used as a correlated image or a reference image of
Needless to say. However, even if a correlated image and a reference image, which are extremely periodic and whose Fourier transform image becomes a delta function or a shear function, are used in the optical pattern recognition device of the present invention, pattern recognition for these deformations can be performed. It has been found that performance is not much improved.
In such a case, it is considered that δm (f ☆ g) or δj (f ☆ g) described in the operation becomes too large to be ignored. However, since such images are scarce except for examples of semiconductor fine processing patterns, the optical pattern recognition method and apparatus of the present invention are considered to be applicable to most pattern recognition applications.

[0092]

As described above, the optical pattern recognition method and the optical pattern recognition apparatus according to the present invention provide an optical pattern recognition method using a matched filter type optical correlator, a joint transform correlator, or frequency filtering of a Fourier transform image. In the method and the optical pattern recognition device, the filter image recorded on the matched filter surface, the joint Fourier transform surface or the Fourier transform surface is separated and recorded into a low frequency component and a high frequency component, and these separated low frequency components are recorded. By independently obtaining identification signals from the filter image corresponding to the high-frequency component and the filter image corresponding to the high-frequency component, and performing pattern recognition using the ratio of these identification signals, the rotation or size of the correlated image or the reference image Optical pattern recognition that is resistant to changes was realized.

This optical pattern recognition method and apparatus uses a practical spatial light modulator such as an optical writing type ferroelectric liquid crystal spatial light modulator or a liquid crystal television, and has relatively small rotation and size change. Because it can perform pattern recognition on objects in real time, it can inspect electronic parts, machine parts, molded parts, etc. on the production line of the factory, collate radiographs and fingerprints, and identify scenery It is very effective in identifying goals. Of course, as shown in the embodiment of the present invention, the effect on the identification of characters and numerals is great.

[Brief description of the drawings]

FIG. 1 is a process diagram showing a flow of an optical pattern recognition method of the present invention based on a joint transform correlator.

FIG. 2 is a process diagram showing a flow of an optical pattern recognition method of the present invention based on a matched filter type optical correlator.

FIG. 3 is a configuration diagram showing an embodiment of the optical pattern recognition device of the present invention based on a joint transform correlator.

FIG. 4 is a configuration diagram showing one embodiment of an optical pattern recognition device of the present invention based on a matched filter type optical correlator.

FIG. 5 is a schematic cross-sectional view showing a configuration of an optical writing type ferroelectric liquid crystal spatial light modulator used in the optical pattern recognition device of the present invention.

FIG. 6 is a graph showing a threshold characteristic with respect to writing light intensity of the optical writing type ferroelectric liquid crystal spatial light modulator shown in FIG.

7 is a graph showing a voltage waveform applied to an optical writing type ferroelectric liquid crystal spatial light modulator in order to realize a threshold characteristic with respect to writing light intensity shown in FIG.

8 is a graph showing a Fourier-transformed image intensity distribution recorded when a Fourier-transformed image is written in an optically-writable ferroelectric liquid crystal spatial light modulator while applying a drive voltage as shown in FIG. 7; is there.

FIG. 9 is a block diagram showing an embodiment of the optical pattern recognition device of the present invention based on a matched filter type optical correlator without using an optical writing type ferroelectric liquid crystal spatial light modulator.

FIG. 10 is a block diagram showing an embodiment of an optical pattern recognition method according to the present invention based on frequency filtering of a Fourier transform image.

FIG. 11 is a diagram showing an example of an input image of the optical pattern recognition device of the present invention based on a joint transform correlator.

FIG. 12 is a diagram showing an example of a low-frequency component and a high-frequency component of a joint Fourier transform image obtained when an alphabet letter B having no rotation or size change is used as a correlated image and a reference image. is there.

FIG. 13 shows a relationship between a correlation coefficient from a high-frequency component, a correlation coefficient from a low-frequency component, and a correlation coefficient ratio with respect to a rotation angle of a reference image when a joint Fourier transform image is binarized. FIG.

FIG. 14 shows a relationship between a correlation coefficient from a high frequency component, a correlation coefficient from a low frequency component, and a correlation coefficient ratio with respect to a change in the size of a reference image when a joint Fourier transform image is binarized. It is a graph shown.

FIG. 15 shows the relationship between the correlation coefficient from the high frequency component, the correlation coefficient from the low frequency component, and the correlation coefficient ratio with respect to the change in the size of the reference image when the joint Fourier transform image is not binarized. It is a graph shown.

[Explanation of symbols]

 Reference Signs List 1 imaging device 2 memory 3 correlated image 4 reference image 5 joint Fourier transform 6 low frequency component 7 intensity distribution image 8 Fourier transform 9 correlation output detection 10 correlation coefficient calculation 11 high frequency component 12 intensity distribution image 13 Fourier transform 14 correlation output Detection 15 Correlation coefficient calculation 16 Correlation coefficient ratio calculation 17 Memory 18 Correlation coefficient ratio 19 Matching 20 Discrimination 21, 22 Fourier transform 23 Carrier wave 24 Fourier transform hologram low frequency component 25 Fourier transform hologram high frequency component 26, 27 Frequency filtering 28 Writing laser 29 First beam expander 30 First CCD camera 31 A / D converter 32 Computer 33 D / A converter 34 Liquid crystal television drive circuit 35 Liquid crystal television 36 First Fourier transform lens 37 Optical writing Type ferroelectric liquid crystal spatial light Adjuster 38 Reading laser 39 Second beam expander 40 Polarizing beam splitter 41 Second Fourier transform lens 42 Second CCD camera 43 Video monitor 44 A / D converter 45 First beam splitter 46 First liquid crystal Television 47 Second beam splitter 48 First mirror 49 Second mirror 50 Second liquid crystal television 51 Third Fourier transform lens 52 First liquid crystal television drive circuit 53 Second liquid crystal television drive circuit 54a, 54b Transparent Substrate 55a, 55b Transparent electrode 56 a-Si: H 57 Dielectric mirror 58a, 58b Alignment film 59 Ferroelectric liquid crystal 60 Spacer 61a, 61b Non-reflective coating 62 Write light 63 Read light 64 First laser 65 Third CCD camera 66 A / D converter 67 Second laser 68 Third liquid crystal display Levi 69 D / A converter 70 Third liquid crystal television drive circuit 71 Condenser lens 72 Reference image 73 Correlated image 74 Correlation coefficient ratio 75 Correlation coefficient from low frequency component 76 Correlation coefficient from high frequency component

──────────────────────────────────────────────────続 き Continuing on the front page (72) Shuhei Yamamoto, Inventor Seiko Electronic Industry Co., Ltd. 6-3-1, Kameido, Tokyo, Japan Sumitomo Cement Co., Ltd. (72) Inventor Yasuhiro Takemura 1 Kanda Midoshiro-cho, Chiyoda-ku, Tokyo Sumitomo Cement Co., Ltd. (56) References JP-A-5-68196 (JP, A) (58) Survey ⁶ (Int.Cl. ⁶ , DB name) G06T 7/00 G02B 27/46 JICST file (JOIS)

Claims (9)

(57) [Claims] 1. An optical pattern recognition apparatus for automatically recognizing or measuring a required pattern by performing an optical correlation process using coherent light on a two-dimensional image obtained from an imaging unit. Means for converting at least one reference image including a required target and at least one newly input correlated image into a coherent image; optically Fourier transforming the coherent image, and performing a Fourier transform of the reference image and the correlated image. Means for independently obtaining converted images; means for independently forming a low-frequency filter mainly composed of low-frequency components of the Fourier transform of the reference image and a high-frequency filter mainly composed of high-frequency components of the reference image; Fourier transform of the correlation image is performed independently by the low frequency filter and the high frequency filter. Means for optically filtering the Fourier transform of the correlated image optically filtered by the low frequency filter and the high frequency filter,
Means for optically Fourier transforming again to independently obtain a first correlation function corresponding to the low-frequency filter and a second correlation function corresponding to the high-frequency filter; and A photodetector that independently detects the intensity of the first correlation peak and the intensity of the second correlation peak included in the second correlation function, and converts the intensity into an electric signal; The intensity of the correlation peak of 1;
Means for calculating a correlation coefficient ratio of the correlated image, which is a ratio of the intensity of the second correlation peak converted to the electric signal, and a reference image, which is an intensity ratio of the correlation peak, from the Fourier transform of the reference image Means for obtaining the correlation coefficient ratio of the reference image in advance and storing the correlation coefficient ratio, and comparing the correlation coefficient ratio of the reference image with the correlation coefficient ratio of the reference image stored in advance to obtain the correlation coefficient ratio. An optical pattern recognition apparatus, comprising: a determination unit configured to determine a correlation image. 2. An optical pattern recognition apparatus for automatically recognizing or measuring a required pattern by performing an optical correlation process using coherent light on a two-dimensional image obtained from an imaging means. Means for converting at least one reference image including at least one target and at least one newly input correlated image into a coherent image; and optically and simultaneously Fourier-transforming the coherent image to joint the reference image and the correlated image. Means for obtaining a Fourier-transformed image, and independently recording or displaying the low-frequency component and the high-frequency component of the joint Fourier transform, and a low-frequency joint power spectrum mainly composed of the low-frequency component of the joint Fourier transform and mainly the joint A high-frequency job consisting of high-frequency components of the Fourier transform A first correlation function corresponding to the low-frequency joint power spectrum, wherein the first and second low-frequency joint power spectrums are respectively subjected to optical Fourier transform independently again. Means for independently obtaining a second correlation function corresponding to the power spectrum and a high-frequency joint power spectrum; an intensity of a first correlation peak included in the first correlation function and a second correlation function included in the second correlation function. A photodetector which independently detects the intensity of the correlation peak of the first correlation peak and converts the intensity into an electric signal; and the intensity of the first correlation peak converted into the electric signal,
Means for calculating a correlation coefficient ratio of a relative correlation image, which is a ratio with the intensity of the second correlation peak converted to the electric signal; and a reference image, which is an intensity ratio of correlation peaks relating to the same two reference images. The correlation coefficient ratio of
Storage means for storing this; and discrimination means for discriminating the correlated image by comparing the correlation coefficient ratio of the correlated image with the correlation coefficient ratio of the reference image stored in the storage means. An optical pattern recognition device, comprising: 3. The means for converting at least one reference image including the required target and at least one newly input correlated image into a coherent image, wherein the total light intensity of the coherent image of the reference image and the correlated image is Are equal,
3. The optical pattern recognition apparatus according to claim 1, further comprising a standardization unit for standardizing the reference image and the correlated image so as to be equal and constant. 4. The means for independently forming the low-frequency filter and the high-frequency filter, or the means for independently forming the low-frequency joint power spectrum and the high-frequency joint power spectrum, respectively, A glass substrate on which a hydrogenated amorphous silicon photoconductive film is formed on a transparent electrode; and a liquid crystal alignment film on each of opposing surfaces of the glass substrate on which the transparent electrode is formed. 2. An optical writing type ferroelectric liquid crystal spatial light modulator in which a pair of formed glass substrates are arranged to face each other and a gap is filled with a ferroelectric liquid crystal composition.
4. The optical pattern recognition device according to any one of items 3 to 3. 5. The means for independently forming the low frequency filter and the high frequency filter, or the means for independently forming the low frequency joint power spectrum and the high frequency joint power spectrum,
Imaging means having a photoelectric conversion surface formed in a matrix,
An analog / digital converter for amplifying an analog image picked up by the image pick-up means and converting it to a digital image;
A computer having a function of masking a predetermined area of the digital image converted by the analog / digital converter, a digital / analog converter for converting the digital image having the predetermined area masked into an analog image, and a digital / analog converter 4. An optical pattern recognition apparatus according to claim 1, further comprising an electric writing type spatial light modulator for displaying an analog image converted by the analog converter. 6. The optical pattern recognition apparatus according to claim 1, wherein said normalizing means comprises a semiconductor laser, a beam splitter, a photodetector, and a liquid crystal mask. 7. An optical pattern recognition method for automatically recognizing or measuring a required pattern by performing optical correlation processing using coherent light on a two-dimensional image obtained from an imaging means. After converting at least one reference image including the target and at least one newly input correlated image into a coherent image, the coherent image is optically Fourier-transformed, and a Fourier-transformed image of the reference image and the correlated image is obtained. Are obtained independently, a low-frequency filter mainly composed of a low-frequency component of the Fourier transform of the reference image and a high-frequency filter mainly composed of a high-frequency component of the reference image are formed independently, and the Fourier transform of the correlated image is performed. Independently optically filtering the low frequency filter and the pre-high frequency filter, Fourier transform of the correlated image optically filtered by the low frequency filter and the high frequency filter,
After optically Fourier-transforming again to obtain a first correlation function corresponding to the low-frequency filter and a second correlation function corresponding to the high-frequency filter independently of each other, the first correlation function included in the first correlation function is obtained. And the intensity of the second correlation peak included in the second correlation function is independently detected and converted into an electric signal, and the intensity of the first correlation peak converted into the electric signal is detected. And a first step of calculating a correlation coefficient ratio of a correlation image, which is a ratio of the intensity of the second correlation peak converted to the electric signal, and a Fourier transform of the reference image, Each of the high-frequency filters is independently optically filtered, and the Fourier transform of the reference image optically filtered by the low-frequency filter and the high-frequency filter is again optically Fourier transformed. 4 corresponding to the third correlation function and the high frequency filter that corresponds to the low frequency filter and
After independently obtaining the correlation functions of the third correlation function, the intensity of the third correlation peak included in the third correlation function and the intensity of the fourth correlation peak included in the fourth correlation function are independently detected, After the conversion into the electric signal, the correlation coefficient ratio of the reference image, which is the ratio between the intensity of the third correlation peak converted to the electric signal and the intensity of the fourth correlation peak, is obtained in advance and stored. And a third step of comparing the correlation coefficient ratio of the correlated image with the previously stored correlation coefficient ratio of the reference image to determine the correlated image. An optical pattern recognition method, comprising: 8. An optical pattern recognition method for automatically recognizing or measuring a required pattern by applying an optical correlation process using coherent light to a two-dimensional image obtained from an imaging means. Converting at least one reference image including the target and at least one newly input correlated image into a coherent image, optically and simultaneously Fourier-transforming the coherent image, and jointly Fourier-transforming the reference image and the correlated image An image is obtained, and the low-frequency component and the high-frequency component of the joint Fourier transform are recorded or displayed independently of each other, and the low-frequency joint power spectrum mainly composed of the low-frequency component of the joint Fourier transform and the high frequency of the joint Fourier transform are mainly obtained. High-frequency joint power spectrum After independently forming the vectors, the low-frequency joint power spectrum and the high-frequency joint power spectrum are again optically Fourier-transformed independently of each other to obtain a first correlation function and a high-frequency function corresponding to the low-frequency joint power spectrum. A second correlation function corresponding to the joint power spectrum is obtained independently, and the intensity of a first correlation peak included in the first correlation function and the intensity of a second correlation peak included in the second correlation function are obtained. Are independently detected, converted into an electric signal, and a specific correlation is a ratio of the intensity of the first correlation peak converted into the electric signal to the intensity of the second correlation peak converted into the electric signal. A first step of calculating a correlation coefficient ratio between images; and a job formed by optically and simultaneously Fourier-transforming the same two reference images. A low-frequency joint power spectrum and a high-frequency joint power spectrum are each formed independently using a Fourier transform, and these are again optically Fourier-transformed independently again, and a third correlation function corresponding to the low-frequency joint power spectrum is formed. And a fourth correlation function corresponding to the high-frequency joint power spectrum are obtained independently of each other. Then, the intensity of the third correlation peak included in the third correlation function and the fourth correlation function included in the fourth correlation function are obtained. After independently detecting the intensity of the correlation peak and converting it into an electric signal, the phase relationship of the reference image is the ratio of the intensity of the third correlation peak and the intensity of the fourth correlation peak converted into the electric signal. A second step in which a numerical ratio is obtained in advance and stored, and a correlation coefficient ratio of the correlated image and a pre-stored Optical pattern recognition method characterized by comparing the correlation coefficient ratio of the reference image and a third step of determining the object correlation image. 9. After the step of converting the at least one reference image including the required target and the at least one newly input correlated image into coherent images, the intensity of the coherent image of the reference image and the correlated image or 9. The optical pattern recognition method according to claim 7, further comprising a normalization step for making the intensity ratio constant.