A window-sequence coding transformation suitable for computationally-efficient bit-serial implementation of stack filters

A window-sequence coding (WSC) technique suitable for improving the computational efficiency of a bit-serial implementation of 2-D stack filters is proposed. The WSC technique takes advantage of the observation that in most images, many pixels appearing in the filter-window at a certain time-instant assume non-distinct values. An algorithm that uses the bit-serial binary-tree search (BTS) architecture for stack filtering and employs the WSC technique is developed. The proposed algorithm is designated as a modified binary-tree search (MBTS) algorithm. It is shown that the computational efficiency of MBTS algorithm for 2-D stack filtering is significantly better than the efficiency of a BTS algorithm employing a recently proposed technique called the input compression. The improvement stems from the fact that in the input compression method, the samples appearing in a filter-window of size M are always mapped to the set of integers (0, 1, ..., M-1), in spite of the fact that usually in a 2-D window, several groups of pixels assume non-distinct values