Chapter 7 - VLSI Computer Arithmetic for Real-Time Image Processing

Abstract

This chapter discusses the application of very large scale integration (VLSI) computer arithmetic for real-time image processing. It describes four partitioned algorithms for modular VLSI implementation of the following four classes of matrix computations: (1) lower–upper decomposition by a new variant of Gaussian elimination, (2) normal inversion of a nonsingular triangular matrix, (3) multiplication of two compatible matrices, and (4) solution of a triangular system of equations by back substitution. The feature extraction and pattern classification are the initial candidates for possible VLSI implementation. The Foley–Sammon feature extraction method and the Fisher's linear classifier have been presented in the chapter for VLSI implementation. Other methods such as the eigenvector approaches to feature selection and Bayes's quadratic discriminant functions should also be realizable with VLSI hardware. It would be advantageous also to develop VLSI computing structures for smoothing, image registration, edge detection, image segmentation, texture analysis, multistage feature selection, syntactic pattern recognition, pictorial query processing, and image database management. The potential merit lies not only in speed gains but also in reliability and cost-effectiveness.