JPH02120917A - Hybrid optical computing element - Google Patents

Abstract

PURPOSE:To rapidly execute both normal and inverse discrete cosine transformation by providing the title computing element with a two-dimensional space optical modulator arranged so that each optical signal corresponding to each picture element of an one-dimensional space optical modulator is modulated by a row or column element. CONSTITUTION:An optical operation part is a part for finding out a prescribed vector matrix product by means of an optical system, an input vector is inputted to the one-dimensional space optical modulator 12 as an electric signal 17 and light with uniform intensity radiated from a light source 11 is modulated and the two-dimensional space optical modulator 14 is irradiated with an input optical signal 18 through an optical system 13 such as a cylindrical lens. Since optical modulation elements with transmittance corresponding to respective element values in a prescribed matrix are set up in the modulator 14, a product result between the input vector and each element of the matrix is obtained as transmitted light. The transmitted light is converged upon an one-dimensional photoelectric converter 16 through an optical system 15 and an electric signal 19 is outputted as an output vector. Consequently, both the normal and inverse discrete cosine transformation can be highly accurately calculated, the processing of an electric circuit can be reduced and rapid operation can be attained.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to an optical hybrid arithmetic unit, and more specifically, it is widely used as an image compression encoding method in still image transmission devices via communication lines. Adaptive discrete cosine transform (A
The present invention relates to an optical hybrid arithmetic unit that performs discrete cosine transform (DCT) and inverse discrete cosine transform (IDCT), which are the main operations in the DCT (DCT) method, at high speed.

[Conventional technology]

The adaptive cosine transform (ADCT) method has high image compression coding efficiency, and the main computational processing is discrete cosine transform (DCT) and inverse cosine transform (IDCT).
) has been discovered and is now used as a coding method for still image transmission. However, in recent years, there has been a demand for faster image transmission equipment due to the increasing definition of images such as high-definition television (HDTV) and the spread of high-speed, high-4W digital communication lines. Because it is necessary to speed up image compression processing, ultra-high speed DCT and IDCT are required.

Conventionally, in order to speed up DCT and IDCTT, in addition to the approach of implementing dedicated LSIs for high-speed algorithms, we have also applied optical arithmetic units, which are known to be able to perform some image processing related to two-dimensional image information at extremely high speeds. Consideration is also being given to doing so. Regarding the application of this optical arithmetic unit, due to the approximation of cosine transform and Fourier transform,
There are many cases in which the optical Fourier transform, which can be easily realized using lenses, is extended and applied.Also, orthogonal transforms such as OCT are originally defined as matrix products, so optical lines that calculate vector-matrix products and matrix products are used. The application of side arithmetic units is also being considered.

[Problem to be solved by the invention]

Among optical calculations, optical Fourier transform is originally a method for processing continuous images, so in order to apply it to DCT, which is a calculation on m-lost images, some preprocessing of the input image and continuous spatial domain are required. An approximation of the DCT is required. As for IDCT, it is difficult to approximate it in a continuous spatial domain from the definition, so it is impossible to realize it by optical Fourier transform. In addition, it is necessary to combine and change the Fourier transform optical system to achieve the desired DCT operation.
There are problems with the optical system turning into dolls and a decrease in calculation speed and accuracy.

On the other hand, the optical side arithmetic unit has an essential problem in that it cannot express and calculate negative numbers. In addition, DCT and IDCT are multistage matrix multiplication operations by definition, but optical side arithmetic units recently developed for real-time calculations of matrix multiplications (for example, IEICE Technical Report ○QE84-122: Real-time multistage Regarding the matrix multiplication optical system), there is a problem that the calculation accuracy is low.

An object of the present invention is to realize a dedicated high-speed arithmetic unit for DCT and IDCT using an optical hybrid arithmetic unit incorporating high-speed optical arithmetic operations.

[Means to solve the problem]

In order to achieve the above object, the optical hybrid arithmetic unit of the present invention divides a two-dimensional image vertically or horizontally in units of pixels, and sequentially connects the row or column vectors I and L to reconstruct it. a one-dimensional spatial light modulator inputting a one-dimensional image;
A spatial filter set with a non-negative light intensity transmittance or reflectance obtained by adding bias transmittance or reflectance to each element value of the coefficient matrix of DCT and IDCT is applied to the light corresponding to each pixel of the one-dimensional input image. Signal is DCT and IDC
further add a row or column consisting of a spatial filter with the same transmittance or reflectance as the bias transmittance or reflectance, corresponding to each pixel of the one-dimensional spatial light modulator; a two-dimensional spatial light modulator arranged such that an optical signal of the output image is modulated by each row or column element, each pixel of an output image reconstructed in one dimension similar to the input image, and the two-dimensional space; a one-dimensional photoelectric converter comprising a photoelectric conversion element corresponding to an additional row or column of a light modulator; a light source that irradiates the one-dimensional spatial light modulator with a uniform amount of light; an optical system that irradiates the two-dimensional spatial light modulator with an optical signal modulated by the spatial light modulator according to the definitions of DCT and IDCT;
The optical signal modulated by the two-dimensional spatial light modulator is subjected to DCT
and an optical system that focuses light onto a one-dimensional photoelectric converter according to the definition of CT.

Furthermore, after photoelectric conversion by the one-dimensional photoelectric converter, the output signal of another photoelectric conversion element is subtracted by the output signal of a photoelectric conversion element provided corresponding to an additional row or column of the two-dimensional spatial light modulator. The present invention is characterized in that it comprises electrical means for removing the bias contribution by removing the bias contribution.

[For production]

In the optical hybrid director of the present invention, DCT and ID
An optical matrix operator is used that can realize both CT operations and does not require conversion processing to a non-essential continuous spatial domain as in the case of optical Fourier transform applications. And D
Since CT, I 2 , and D CT are multi-stage matrix operations, the definition formula is modified so that the optical system can be executed using the simplest one-stage vector matrix operation. In addition, for negative number expression and calculation, the light transmittance or reflectance of the spatial light modulator corresponding to the constant matrix is set based on a bias value, and the bias component is extracted from the final calculation result by electrical means. Remove. This bias component can be obtained as an independent signal in parallel with the DCT and IDCT operations by only slightly changing the optical arithmetic unit, and its removal can be realized by a well-known electronic subtraction circuit.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of an optical hybrid computing device according to the present invention, in which 1 is an optical computing section, 2 is a bias removal circuit, and 3 is a control section. Here, the operation of the pilot hybrid arithmetic unit will be explained using an example of a light transmission type DCT operation.

First, a procedure for converting DCT and IDCT from multi-stage matrix product to single-stage vector matrix product will be described.

The following equations (7) (1) and (2) are the definition equations of DCT and IDCT, respectively.

Now, the secondary image (f (j, k) I jl k:o,
1. -, N-1) in the row direction and sequentially connect them.') Ll (f ''(Q) I Q=
0.1. −, N”−1) h, (F(up v) l
Similarly, Ll, v:0+ 1+ . . . , N-1) is transformed into a row vector (F" (w) l w=0.1.-, N"
-1) F, depending on the distortion (4C(u)C(v) ・
cos [(2j+1) uπ/ 2 N]・cos
[(2 + 1) V 7C/ 2 N] /N21
Jl kr u, v=o.

1, -=, N-1) as a matrix (g(R,%I) IQ,
v=o, 1゜・=, N”-1) h, (C(u)
C(v) ・cos [(2j+l) uπ/ 2 N
] ・cos [(2 + 1) V7C/ 2 N]
F+ L up V=O, l, -, N-1)
matrix entrance 1 (R1, w) I Q , w:o.

1, -, N"-1). That is, by setting Q = j 10 kN and w = u + νN, it can be transformed into the following back 1 - le matrix product. Equations (3) and (4) are Each//DC
"I', IDCT.

However, C(u)=1/v'2 (u=O), 1 (
u=1,2. -N-1), O(u=N, N+1.-) where f(j, k) is a two-dimensional image and F''(u, v) is the DCT power spectrum of that image.

The optical calculation unit 1 shown in FIG. 1 is a part that calculates the above-mentioned vector matrix product using an optical system. FIG. 2 shows a specific example of the configuration of the optical matrix calculator. According to Figure 2? The light source 11 and the one-dimensional spatial light converter 12 are one-dimensional LED arrays, and the two-dimensional spatial light modulator 14 is a metal plate with an open area corresponding to the set transmittance. The transducer 16 made by the company is realized by a linear image sensor.

The operation of the optical power calculator will be explained using DCT calculation as an example.

Input vector (f"(Q)l Q=0.1...,
N''-13 is input to the one-dimensional spatial light modulator 12 as the upper air signal 17, and the one-dimensional spatial light modulator 12 modulates the light of uniform intensity from the light source 11, so that the input optical signal 18 is The optical signal 18 is generated by an optical system 13 such as a cylindrical lens that can focus light only in the X-axis direction.
The beam is expanded into a parallel beam with uniform light intensity in 11 directions, and is irradiated onto the two-dimensional spatial light modulator 14.

The two-dimensional optical modulator 14 includes a matrix (gcQ, ν)Q, do 0, 1 . -..., N2-1), the light modulation elements with transmittance corresponding to each element value are set in an arrangement with rows in the X-axis direction and five columns in the y-direction, and calculations are made for each element of the input vector and matrix. Result f"(Q)・g(Q,,w) is obtained as transmitted light.": The transmitted light is transmitted to a photoelectric converter 16 by an optical system 15 such as a cylindrical lens that has the ability to focus only in the X-axis direction. The total light intensity is detected. The photoelectric converter 1G outputs an electric signal 19 corresponding to the focused optical signal as a vector (F"(W)1=o, 1..., N"-1)
Output as .

Although negative calculations cannot be performed with an optical power calculator like the one shown in Figure 2, 1
Matrix (gcQ, omitted) 1Ω, do 0, 1. ..., N2-
1) also includes negative elements. Bias removal circuit 2 in Figure 1
was added to solve this problem. FIG. 3 shows the relationship between the optical calculation section 1 and the bias removal circuit 2 in that case. Here, the bias removal circuit 2 includes a scanning circuit 21 and a subtraction circuit 22.

Next, referring to FIG. 3, it will be explained that both forward and reverse DCT operations are achieved.

The two-dimensional spatial light modulator 14 is expanded to N+1 rows, and a transmittance t(Q, ν) determined by the following formula is set for each element.

t (Q, w) = a ・K (Q , w) + β (w =
o, 1. ..., N-1) (5) β
(w=N) However, α is a coefficient to match the transmittance dynamic range of the two-dimensional spatial light conversion adjuster 14, and β is t (U, W
) to make it non-negative.

2\, the optical signal 18 from the one-dimensional spatial light modulator 12 is
(Q)=γ・f”(Q) (γ is the conversion coefficient to light intensity), and if we ignore the loss on the way, one-dimensional photoelectric converter 1
The intensity of the optical signal incident on the optical signal 6 is expressed by the following equation, ignoring the loss on the way.

(Lee N) The scanning circuit 21 of the bias removal circuit 2 receives an output from the one-dimensional photoelectric converter 16. Of the electrical signals corresponding to (w), the parts o (0) to o (N''-1) are scanned and sequentially input to one side of the subtraction circuit 22. The input is o (N). With this N,
The sum of the bias components is.

(N), so the subtraction circuit 2
By subtracting o(0)-o(N''-1).(N) in 2 and removing the bias component, the desired signal 2 can be easily obtained.
You can get 3.

The control unit 3 in FIG. 1 sequentially reads sub-blocks to be subjected to DCT calculation from an external image storage means via an interface 31 under the control of a still image transmission device or the like in which the optical hybrid DCT calculation unit is incorporated, and converts them into analog It is converted into an electrical signal 17 and sent to the optical calculation section 1. The calculation result of the optical calculation unit 1 is input as an electric signal 19 to a bias component removal circuit 2, the bias component is removed, and the result is output as an electric signal 23. The control unit 3 sequentially takes in the output electrical signal 23 until the processing of -images is completed, converts it into a digital electrical signal, and stores it in the image storage means via the interface 31.

Above, we have explained the light transmission type DCT operation as an example, but IDCT
And the reflective type can also be easily inferred.

〔Effect of the invention〕

As described in detail above, according to the present invention, an optical hybrid arithmetic unit is capable of calculating both forward and inverse discrete cosine transforms with high arithmetic precision, and is also capable of speeding up processing by reducing processing in electronic circuits. can be realized.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a detailed diagram of an optical matrix arithmetic unit, and FIG. 3 is a diagram showing the relationship between the optical arithmetic unit of FIG. 1 and the bias component removal circuit. It is. 1... Light calculation section, 11... Light source. 12... One-dimensional spatial light modulator, 13... Optical system, 14... Two-dimensional spatial light modulator, 15... Optical system, 16... One-dimensional photoelectric converter, 2... Bias component removal circuit, 21...scanning circuit,
No.lr.

Claims (3)

[Claims] (1) A one-dimensional spatial light modulator that inputs a one-dimensional image obtained by dividing a two-dimensional image vertically or horizontally into pixel units and sequentially connecting the row or column vectors, and a discrete cosine transform. and a spatial filter set with a non-negative light intensity transmittance or reflectance obtained by adding a constant bias transmittance or reflectance to each element value of the coefficient matrix of the inverse discrete cosine transform, corresponding to each pixel of the one-dimensional input image. Arrange the optical signal to be modulated according to the definitions of the discrete cosine transform and inverse discrete cosine transform, and add a row or column of spatial filters with the same transmittance or reflectance as the bias transmittance or reflectance, and a two-dimensional spatial light modulator arranged such that an optical signal corresponding to each pixel of the original spatial light modulator is modulated by one row or column element, and one-dimensionally reconstructed in the same manner as the input image; a one-dimensional photoelectric converter comprising a photoelectric conversion element corresponding to each pixel of an output image and an additional row or column of the one-dimensional spatial light modulator; and a light source that illuminates the one-dimensional spatial light modulator with an even amount of light. and an optical system that irradiates the two-dimensional spatial light modulator with an optical signal emitted from the light source and modulated by the one-dimensional spatial light modulator in accordance with the definitions of discrete cosine transform and inverse discrete cosine transform; An optical hybrid arithmetic unit comprising: an optical system that focuses an optical signal modulated by a spatial light modulator onto the one-dimensional photoelectric converter according to definitions of discrete cosine transform and inverse discrete cosine transform. (2) After photoelectric conversion by the one-dimensional photoelectric converter, subtracting the output signal of another photoelectric conversion element with the output signal of the photoelectric conversion element provided corresponding to the additional row or column of the two-dimensional spatial light modulator. 2. The optical hybrid arithmetic unit according to claim 1, further comprising an electric means for removing said bias contribution. (3) It has an interface with an external image information storage means, and supplies an input image signal to the one-dimensional spatial light modulator, and also takes in the output signal from the electric means to perform discrete cosine transformation and inverse processing. 3. The optical hybrid arithmetic unit according to claim 2, further comprising a control unit that completes the discrete cosine transformation and outputs the result of the calculation to the image information storage means.