Investigation Of Matched Filers Design For Optical Pattern Recognition

Optical pattern recognition, which is based on optical Fourier transform, with high-speed operation, parallel processing and not subject to input image size compared to traditional digital pattern recognition methods, has a wide range application in military, industry and other fields.Matched filters are limited by the sensitivity of their correlation peaks to size and angle changes of reference target. In this thesis, the design, optical display and correlation experiment of matched filters in Vander Lugt correlators are studied to resolve the problem of distortion-invariant pattern recognition. The main contents of the study are as follows.Firstly, a photographic model is established to quantitatively analyze the impact of scale zooming, in-plane rotation, perspective angle variation and gray distribution on correlation output,as a result, the distortion intervals of training set images are choosed.Secondly, the characteristics of optimal trade-off synthetic discriminant function (OTSDF) filter including the training set images, parameters selection, discrimination performance, noise and clutter rejection are discussed based on some preliminary synthetic discriminant function filters. A searching and management method of optical filter is introduced to achieve 4-classes target recognition covering 0~360°perspective angle variation. Minimum background energy OTSDF filter is proposed to resolve the problem of low discriminant ratio in clutter background, it can reject background noise effectively and the correlation output has high signal to noise ratio. On this condition, the correlation and matched filters design of targets in clouds and ground environment are studied.The space-bandwidth product of spatial light modulators (SLM) and optimal filters is always inconsistent, a solution to this problem is using the minimum Euclidean distance principle to implement ideal filter to SLM. We introduce a single matched filters implementation method and dictionary searching method for OTSDF filter implementation based on the measuring of the phase modulation characteristics of SLM. A 2f correlation system is built to carry out optical pattern recognition, and the experiment results testify the OTSDF filter design and implementation methods.Finally, the system error model of correlation pattern recognition is established, the relationships of reposition error, partial matched error and position error with correlation output are presented according to the simulation results.