VLSI ARCHITECTURE FOR THE IMPLEMENTATION OF REAL-TIME ORDER STATISTIC FILTERS

摘要

A VLSI ARCHITECTURE FOR THE IMPLEMENTATION OF REAL-TIME ORDER STATISTIC FILTERS N. Rama Murthy and M.N.S Swamy Centre for Signal Processing and Communications Department of Electrical and Computer Engineering Concordia University 1455 de Masionneuve Blvd. W. Montreal, Quebec H3G 1M8 This invention relates to the real-time implementation of a class of nonlinear digital filters known as Order Statistic Filters. These type of filters have found application in several areas of signal and digital image processing such as the processing of pictures obtained from satellites, x-ray and other sources. Real-Time implementation of order statistic filtering as used in signal and image processing involves computing, at every sample output instant, the ranked order of N input signal samples in a window which gets periodically updated with the arrival of a new input sample, where N is usually odd and is assumed to be so from now on. Let xp(n),p=1,2,...,N be the samples inside the window wN(n) at the n th time instant, i.e., wN(n)={x1(n),x2(n),...,xN(n Define a new window wN as the one obtained from wN(n) by sorting the samples xp(n), p = 1,2,..., N inside wN(n) in the increasing order of algebraic value, i.e., wN(n) = {x(1)(n),x(2)(n),...,x(N)(n) } where x(1)(n) ? x(2)(n)?...? x(N)(n). The output Y(n) of an i th Order Statistic Filter (OSF) with input wN(n)is x(1)(n). In the case of one dimensional order statistic filtering, wN(n) gets continuously updated at every sampling instant n with the deletion of the oldest sample (in time) and the inclusion of the latest sample (current or new sample). In the case of 2-D median filtering of digital images, a two dimensional mask of size N1 x N2(where usually both N1 A N2 are odd) centered on a pixel x(u,v) being processed at instant n, is used to select N = N1 ?N2 samples of wN(n). In this case, as the mask is moved along the digital image for processing of all the pixels in the image, more than one old sample gets deleted from wN(n) and an equal number of new samples get included in wn(n). In practice, the signal samples that constitute wN(n) are obtained from the output of an A/D converter used for quantization and coding. If the largest possible sample inside wN(n) at any instant has r-bit binary representation, then real-time order statistic filtering involves computation of the r-bit orderstatistic at every sampling instant (running order statistic) which is a computationally challenging task to implement in real-time for the processing of video signals.