Online pattern recognition in noisy background by means of wavelet coefficients thresholding

Abstract

When a scene contaminated with noise has to be recognized by using an optical correlator, the output plane may be strongly affected by noise, yielding mistaken results. Different strategies have been developed to overcome this problem. In this paper we present a method that applies the thresholding of the wavelet coefficients to perform recognition tasks with scenes contaminated with additive noise. The method is implemented by using a Vander Lugt correlator architecture operating with liquid crystal displays. A unique filter is designed to accomplish the recognition and the denoising processes in a simultaneous way. The function to be recognized is decomposed into sub-bands based on the Gabor decomposition, in the frequency plane. The hard thresholding operation is performed and the threshold is generated with accurate support functions in the filter plane. The criterion for the threshold selection is chosen to optimize the signal-to-noise ratio in the output plane. As examples, numerical simulations and experimental results for the correlation plane are shown. We also compare some quality parameters with classical filters.