SYSTOLIC STRUCTURES FOR TEXTURE REPRESENTATION AND ANALYSIS

Abstract

This thesis proposes systolic implementations for the processes of texture representation, and texture distance computation based on the transportation simplex algorithm. The systolic structure for texture distance computation consists of a two-dimensional array of cells whose size depends on the number of discrete events used to represent texture patterns in terms of a given property. In addition, this array is surrounded by peripheral cells which are storage areas for the data required by the systolic structure for its computation. Systolic structures for applying the Sobel operator to texture images and to obtain frequency histograms are also proposed. Using these structures, the computational time for texture distance computation is 0(n2), and that for texture representation is 0(n). In both cases, there is an increase in speed of 0(n) over corresponding sequential approaches with no parallel processing capabilities.