The Experiment Research On Optical-electrical Hybrid Joint Transform Correlator

The optical pattern recognition has the advantages of being with superior high speed and parallel processing, which makes it a very important way to realize pattern recognition in real- time. However, its disadvantage is of low precision. On the other hand, the digital pattern recognition has the advantages of being with high precision, flexible storage, easy estimation and control, but it has disadvantages, including massive storage requirement, and the slow processing speed. Based on the combination of these two methods, a feasible optical-electrical hybrid joint transform correlator is proposed in this paper.Joint transform correlator (JTC) is usually used in optical image recognition systems. The correlative theories of joint Fourier transform and the structure of the optical cross-correlation is introduced in detail. The quick development of liquid crystal display (LCD) brings opportunity to simplify the optical inspection system.In the target recognition system based on JTC, The autocorrelation peaks of target image and reference image display overlapped in the center of output plane. Its high-intensity and wide-breadth limit the size and relative position of target and reference images, with the influences of stray light, the cross-correlation peaks we want probe would be drowned out. So we need to pretreat the joint image and power spectrum before sending them into LCD, so the recognition capability of JTC can be uprated. The traditional pretreatment include edge detection, binaryzation, sharpening and so on. All these methods need be operated by computer, but pretreatment of real-time collected images by calculator will consume some time, not fit for the JTC operated in real-time pattern.In the experiment, we proposed a new method of power spectrum processing, the traditional pretreatment methods, such as binaryzation or sharpening, are not used again. Only a Gaussian high-pass filter is put on the spectrum, so the treating speed is relatively fast. This method can weaken the zero-order bright spot in the correlation plane, narrow the bandwidth of the zero-diffractive peak, leach noises efficaciously and sharpen the correlated peak. Compared with binaryzation of power spectrum, the method can avoid large losing of useful details of the image (high-frequency signals).We do experiments of angle recognition basing the Gaussian mask method and get good results, further proving the simpleness and availability of Gaussian mask method.