GB2219179A - Coherent optical processing apparatus - Google Patents

Abstract

Two input images are formed from arrays (Figs 2, 3) of pixels by arrays of electro-optical transducers (12) such as SLM's illuminated by a laser (10) and collimator (11), each array of pixels comprising sub-arrays corresponding to different bit weights. The sub-arrays of one array are time-domain modulated in sequence by an array of shutters (13). The sub-arrays of the other array are amplitude modulated, again in accordance with bit weights, by an array of attenuators. A filtering operation is implemented by a filter (16) in the Fourier plane between two lenses (15, 17) and the filtered image is reconstituted by arrays of detectors (19), e.g. CCD's and an output processor (25) which sums the weighted bits pixel by pixel. After each shutter (13) has been opened a Fourier plane image has moreover been written into a non-linear device (21) with memory, via another lens 20. This image is read out by another laser (27) of different wavelength, collimator (28) and lens (22) to form convolution and cross-correlation functions at an output plane. An array of detectors (23) at this plane detects the cross-correlations, between the sub-array selected by the shutter (13) and the sub-arrays of the said other array. The detectors (23) e.g. CCD's integrate the cross-correlations over the sub-arrays of the said one array and an output processor sums the weighted bits pixel by pixel to produce an image representing the correlation between the two input images. <IMAGE>

Description

COHERENT OPTICAL PROCESSING APPARATUS The present invention relates to coherent optical processing apparatus for processing multi-bit data and is concerned particularly but not exclusively with apparatus for filtering data and for correlating data by obtaining the correlation between two input functions represented by images in an input plate. The principles of apparatus for forming correlations between two images are well known and are described for example in "Basic Concepts" D. Casanent and H.J.Caulfield, Optical Data Processing, Topics in Applied Physics, Springer Verlag, 1978 pp 1-16.

Further information is given in "Coherent Optical Processing", S.H.Lee, Optical Information Processing, Springer Verlag, 1981, pp 43-68.

The advantage of coherent optical processing is that it can handle operations which are computationally very difficult because of the data rate. In known applications it is assumed that the input functions are analog. The object of the present invention is to provide apparatus which can be used to operate a multi-bit data and, in one specific application, form correlations between multi-bit digital input functions.

Fig. 1 is a schematic perspective view of apparatus embodying the invention, Fig. 2 is a plan view of an input spatial light modulator (SLM), Fig. 3 is a plan view of an attenuator array, Fig. 4 shows the diffraction pattern of a single pixel subarray, Fig. 5 illustrates a filter function, Fig. 6 illustrates images simultaneously present in an input plane, and Fig. 7 is a plan view of a detector.

Figure 1 shows a laser 10 and optical collimator 11 which produces an expanded beam of coherent light. The light is directed through an input SLM 12 and then through an attenuator array 13, although the order of these elements is immaterial. The SLM 12 establishes the two input images, as explained below and the attenuator array 13 weights the bits of the two images, again as explained below. The SLM 12 and attenuator array 13 are controlled by a unit 14.

The images in the input plane are Fourier transformed by a first lens 15 and a filter 16 in the Fourier plane performs filtering in the spatial frequency domain. A second lens 17 transforms back to the spatial amplitude domain and the filtered images are reflected by a beam splitter 18 on to a detector array 19 which provides, for each input images, an analog signal for each pixel of the filtered image.

The light passing through the beam splitter 18 is converted back into the spatial frequency domain by a third lens 20. A non linear element 21 in the Fourier plane, a read-out laser 27 and collimater 28 and a fourth lens 22 form correlations in an output plane, wherein there is provided a second detector array 23.

The SLM 12 (Fig.2) comprises two arrays 30 and 31 handling the two input images quantized to 12 bits. Each array 30, 31 comprises 12 sub-arrays 30(0)-30(11) and 31(0)-31(11) arranged in two columns and six rows for handling the zero'th to eleventh bits respectively of the quantized images. Each sub-array finally comprises m x n pixels in general, n2 pixels in the case (hereinafter assumed) in which the images are dissected into square n x n pixel arrays. The value of n is chosen according to the required image size to be processed. Each sub-array 30(0)-30(11), 31(0)-31(11) is square with a side a. The sub-arrays are spaced by a and the two major arrays 30 and 31 are spaced by 5a.

The spatial light modulator 12 thus establishes a total of 2 x 12 x n2 pixels, each of which may be individually turned on (transmissive) or off (blocked) by the control unit 14 of Fig. 1.

The control unit receives image data, e.g. as a pixel-serial, bitparallel data stream from each of two quantizers handling the two input images and feeds the 2 x 12 pixel-serial data streams to drivers individual to the 2 x 12 sub-arrays 30(0)-30(11), 31(0)31(11).

The attenuator array 13 (Fig.3) comprises an array of 12 shutters 40(0)-40(11) registering with the first image sub-arrays 30(0)-30(11) respectively. The shutters are implemented as further SLMs which are not controlled on a pixel by pixel basis. Rather the control unit 14 renders each shutter in turn wholly transmissive, but for a sequence of times which may follow a binary geometric progression.- The shutter 40(0) for the LSB b(o) is opened for a very short interval t, e.g. 10 ns. Assuming a binary geometric progression, the shutter 40(1) for the next bit b(l) is opened for 2t and in general the shutters are opened for a time Ti = t.2(i + 1) where i is the bit index. The integrated amounts of light transmitted via transmissive pixels are thus weighted in accordance with the bit weights of the sub-arrays 30(0)-30(11).

In practice it is preferred to open each shutter for slightly longer than the time Ti and, while the shutter is open, to pulse the laser 10 for exactly Ti - hence the connection shown in Fig.1 from the control unit 14 to the laser - since it is possible to control the laser 10 more precisely than SLM devices.

Returning to Fig. 3, the attenuator array 13 further comprises 12 fixed attenuators 41(0)-41(11) registering with the sub-arrays 31(0)-31(11) respectively. The fixed attenuators are permanently transmissive but have transmission factors decreasing, e.g. by factors of 2 from the attenuator 41(11) down to the attenuator 41(0), as indicated by the increasingly heavy shading.

The time domain and amplitude domain attenuation implemented by the attenuator array 13 is required to weight the bits correctly for the joint transform processing performed to determine the correlation and also for the generation of the filtered images at the detector array 19. So far as the latter generation is concerned it is immaterial whether time domain or amplitude domain attenuation is utilized but this combination of the two forms of attenuation is crucial in determining the correlation, as will be explained below.

Fig. 4 shows the main features of the diffraction pattern 2.

formed by a single n pixel sub-array in the Fourier plane of the first lens 15. A zero-order blob 50 at the origin is surrounded by rings of weaker and weaker blobs on the x and y axes and even weaker blobs (not shown) off the axes. The information content of the pixel array is all present within the circle 51, essentially within the zero order blob, (although it is repeated in outer rings) and the first order blobs 52 arise from the structure of the gaps between the pixels. Each blob actually has fine structure arising from the fact that there are 24 sub-arrays contributing to the diffraction pattern. However the information content of all subarrays is present within the circle 51 and it is therefore possible to apply the same filter function to all bit planes of both input images by a single filter 16 in the Fourier plane of the lens 15.

In fact the higher order blobs actually add the information related to the pixel and inter-pixel geometry, which information is not required. Accordingly the filter 16 is used to cut out the unwanted information and retain that which relates to the gross structure of each sub-array, i.e. which pixels are l10sl and which pixels are "1", this being the information representing the input images.

The precise characteristics of the filter 16 are not material to the present invention but it is apparent that they must represent the Fourier transform of the amplitude and/or phase filter characteristics desired in the image plane of the SMD 12. These may, as one example, be those of a Wiener filter, in which case the filter characteristics in the spatial frequency domain are a kind of two dimensional Gaussian function of the form illustrated in Fig.5, in which the x and y axes correspond to those of Fig.4 and the z axis represents transmission factor. This filter function extends essentially over the area of the circle 51 (Fig.4) in the Fourier plane of the lens.

The filtered images reconstituted by the second lens 17 are projected by that lens and the beam splitter 18 on to the detector array 19, which is not separately illustrated since it appears 2 schematically the same as Fig.2, except that the 2 x 12 x n SLM pixels of Fig. 2 become CCD cells. Accordingly the sub-arrays of Fig.2 can now be regarded as representing detector CCDs.

Considering the first image, each COD 30(0)-30(11) integrates the light falling thereon during the time that the corresponding shutter 40(0)-40(11) is open. After all 12 light pulses have been integrated the 12 CCDs store their respective bit planes of the filtered image, weighted in accordance with bit values. The outputs of the 12 CCD's can be summed electrically in a first output processor 25 (Fig.1) to provide a pixel by pixel analog representation of the filtered first image.

The situation is essentially the same for the second image.

However all sub-arrays receive the same total exposure t + 2t + 4t + .......+ 2" t since there is no shuttering. However the detector CCDs 31(0)-31(11) receive light via the fixed attenuators 41(0)-41(11) so that weighting in accordance with bit values is again established.

Turning now to generation of the correlation, it is known that a non-linear optical element with memory (e.g. a BSO crystal) placed in the Fourier plane can create in the output plane the convolution and correlation functions of two input plane functions (Lee, Fig.2.15 and description thereof), when read out using light of a different wavelength from that used to write the information in the crystal The present invention extends this concept to forming 12 cross-correlations simultaneously, one per bit plane of the second image, and doing this sequentially for all 12 bit planes of the first image.

Let the first image be

and the second image be

in which i is the bit plane index and the functions q and q' represent quantizing noise. The bits planes Fi(x,y) and Hi(x,y) may be written more simply FO - F11 and HO - H11.

Fig. 6 illustrates the images present in the input planes of the lenses 15 and 20 when, as an example, the F9 bit plane shutter 40(9) is open. There are 13 images in the input plane and a multiplicity of correlations and convolutions will be created by the non-linear element 21 with other unwanted outputs (analagous to the unwanted outputs of a signal mixer). In order to separate the desired functions from the undesired functions, a reference delta function is also incorporated in the input plane, as explained by Lee, which additionally illuminates the non-linear device with a tilted plane wave and biases this device to the desired operating point.

The write laser 10 (Fig.l) may be a green laser, the read laser 27 a red laser. After each shutter 40(i) has been opened for its respective length of time Ti, the read laser 27 is pulsed for a fixed length of time to irradiate the non-linear crystal 21 uniformly via the beam splitter 18 and thereby form the corresponding correlation and convolution functions at the output plane.

Again considering the example of Fig. 6, namely i = 9, a single set of 12 cross-correlations is in fact selected as the desired output, namely the cross-correlations * * * F9*HO , F9*H1 k F9*Hll and these cross-correlations fall on to an array of 12 CCDs 60(0)- 60(11) shown in Fig.7 and constituting the detector array 23 of Fig.l. Each CCD has n2 cells. The set of cross-correlations obtained after the F9 shutter 40(9) has been opened can be collectively denoted C(9). In a full sequence of 12 laser pulses with the shutters 40(0) to 40(11) opened in turn and the laser 27 subsequently pulsed for the same time on each occasion, 12 correlation sets C(O) to C(11) are obtained. These are all integrated by the 12 CCD's 60(0) to 60(11).At the conclusion of the cycle, the 12 bit planes of the complete correlation function of the two (filtered) input images are held in the 12 CCD's. The outputs of the CCD's are added electrically in a second output processor 26 (Fig.l) to provide the pixel analog values representing the correlation over the planes of the two images.

The two input images may or may not have pictorial significance. They are created by the two 12-bit data streams input to the control unit 14 and the physical entity represented by the data streams is not necessarily a pictorial image. 12-bits have been selected as an example. The invention could obviously be used with fewer or more bits and, as already indicated, the number of pixels processed (n for a square array) is chosen as desired.

The light used in practising the invention is not necessarily visible light.

The invention is not limited to specific details of the embodiment as described above. SLM's are commercially available electro-optical transducers but other possible transducers are liquid crystal devices, magneto-optic devices and Heriot-Watt devices. The shutters in the attenuator array 13 are described as SLM's but may likewise be formed by other kinds of optical switches.

Equally CCD's are commercially available detectors (opto-electric transducers) but other detectors can be used. Lenses are shown as simple lenses. In practice detailed optical design is likely to lead to the use of compound lenses.

If the outputs provided by the first detector array 19 are not required, this array may be omitted along with the second and third lenses 17 and 20. The non-linear element 21 is placed in the Fourier plane of the first lens 15, in series with the filter 16 if this is required.

Claims (16)

1. Coherent optical processing apparatus for processing a plurality N of items of data, each of a plurality B of bits, comprising input means for forming an array of B coherent input light images, each composed of N pixels representing N bits of equal significance from the N items of data respectively, first lens means operative on the array of light images to form in the Fourier plane of the lens means a pattern of light energy in the spatial frequency domain, optical means for modifying the said pattern of light energy, second lens means converting the modified pattern of light energy back to an output image in the spatial amplitude domain, and detector means responsive to the output image.

2. Apparatus according to claim 1, wherein each input light image is an n x m rectangular array of pixels where B = n x m.

3. Apparatus according to claim 1, wherein each light image is an n x m square array of pixels, where B = n2

4. Apparatus according to claim 1, 2 or 3, wherein the optical means is a filter.

5. Apparatus according to any of claims 1 to 4, comprising means for weighting the contributions of the input light images to the said pattern of light energy in dependence upon the corresponding bit weights.

6. Apparatus according to claim 5, wherein the weighting means comprise an array of B shutter devices operable sequentially to allow light to pass through the first lens from the input light images one by one in turn and for durations depending upon the corresponding bit weights.

7. Apparatus according to claim 6, wherein the shutter devices are opened for over-long durations, during which a laser forming the source of coherent light is pulsed on for the precise durations which depend on the bit weights.

8. Apparatus according to claim 5, wherein the weighting means comprise an array of B attenuators in series optically with the B input images respectively.

9. Apparatus according to any of claims 1 to 8, wherein the detector means comprise an array of B detectors corresponding to the B bits respectively, each deflector having N detector cells corresponding to the N pixels respectively.

10. Apparatus according to claim 9, wherein the detector cells produce outputs representative of the integrated light energy impinging thereon over a period of time.

11. Apparatus according to claim 9 or 10, comprising summing means for summing signals from the B detectors pixel by pixel.

12. Apparatus according to claim 1, 2 or 3, wherein the input means are responsive to two pluralities of N items of data to form first and second spaced arrays of B coherent input light images, each composed of N pixels, first weighting means comprising an array of B shutter devices operable sequentially to allow light to pass through the first lens from the input light images of the first array one by one in turn and for durations depending upon the corresponding bit weights, second weighting means comprising an array of B attenuators in series optically with the B input images respectively of the second array, and wherein the optical means comprises a non linear element adapted to form an output image in response to each shutter device being open which output image represents a set of crosscorrelations between one input light image of the first array and all input light images of the second array.

13. Apparatus according to claim 12, wherein the non-linear element has memory and the optical means further comprises means for illuminating the non-linear element with light having a different wavelength to that of the input light images, after each shutter device has been open.

14. Apparatus according to claim 12 or 13, wherein the detector means comprise an array of B detectors corresponding to the B bits respectively, each having N detector cells corresponding to the N pixels respectively and on to which impinge in turn the B sets of B cross-correlations produced as the B shutter means are operated in turn.

15. Apparatus according to claim 14, wherein the detector cells produce outputs representative of the integrated light energy impinging thereon over the total durations of operation of the shutter devices.

16. Apparatus according to claim 15, comprising summing means for summing signals from the B detectors pixel by pixel after completion of integration over the total durations of operation of the shutter devices.