A MODIFIED BINARY-TREE SEARCH ARCHITECTURE FOR 2-DIMENSIONAL STACK FILTERING

A modified binary-tree search (MBTS) architecture suitable for the dev elopment of a computationally efficient implementation of 2-D stack fi lters for image processing applications, is presented. The MBTS archit ecture uses a binary-tree search (BTS) configuration for stack filteri ng in conjunction with a new technique for encoding the samples appear ing in the input window of the stack filler. This encoding technique i s designated as a window-sequence coding (WSC) transformation, which t akes advantage of the observation that in most images, many pixels app earing in the filter-window at a certain time instant assume non-disti nct values. A comparison between the computational efficiency of a BTS -based architecture employing a recently developed encoding technique called input compression, and the efficiency of the proposed MBTS appr oach for performing an operation of 2-D stack filtering, is carried ou t. The comparison illustrates that with the MBTS approach, significant improvements in terms of the computational time can be achieved, with out an increase in the chip area. (C) 1997 Elsevier Science B.V.