Application of a class of parallel processing architecture for implementation of image processing and, more specifically, image pattern recognition is proposed. Noting the problem of optimally matching a wide variety of possible processor configurations to a broad spectrum of different algorithms, a general framework for characterizing image processing algorithms is proposed, and the task of mapping the algorithm characteristics into processing system characteristics is begun. A simple image pattern recognition method is considered from this point of view. An algorithm is developed and alternative SIMD (single instruction stream - multiple data stream) parallel implementations are compared. Complexity analyses are presented to show the computational speedup made possible by the parallelism.