Parallel Implementations of Recognition Techniques

Abstract

Tasks in image processing and computer vision are computationally very intensive. Furthermore, many applications such as robotics, target acquisition, and navigation require real time or near real time processing. Thus, developing efficient algorithms matched to high-performance architectures is a critical area of research in computer vision. In this chapter we describe parallel implementations of the object recognition algorithms described in Part II. The discussion in this chapter is very limited in scope, mainly discussing our work of parallel implementation of algorithms given in Part II using two different parallel architectures, the Connection Machine (CM) 1 and the hypercube. For general discussions of parallel computing in vision, the interested reader could refer to Duff and Levialdi [1981], Duff [1978, 1986], Levialdi [1985], Reeves [1984], Uhr [1984, 1987a, 1978b], Yalamanchili et al. [1985], Cypher and Sanz [1989], Chaudhary and Aggarwal [1990], Crisman and Webb [1991], and Preston and Uhr [1982]. For a general reference for parallel architectures, Almasi and Gottilieb [1989] and Hwang and Briggs [1984] are excellent references.