SIMD Tree Algorithms for Image Correlation

This paper examines the applicability of fine-grained “pure” tree SIMD machines, which are amenable to highly efficient VLSI implementation, to image correlation which is a representative of low-level image windowbased operations. A particular massively parallel machine called NONVON is used for purposes of explication and performance evaluation. Several algorithms are presented for image shifting and correlation operations. Novel algorithmic techniques are described, such as vertical pipelining, subproblem partitioning, associative matching, and data duplication that effectively exploit the massive parallelism available in finegrained SIMD tree machines. Limitations of SIMD pure tree machines are also addressed. They tend to correspond to situations in which the root of the tree may become a significant communication bottleneck, or in situations in which MIMD techniques would be more effective than the SIMD approaches considered in this paper. Performance results have been projected for the NON-VON machine (using only its tree connections, in order to address the issues of concern in this paper).