MORPHOLOGICAL OPERATIONS ON MESH CONNECTED ARCHITECTURE: A GENERALIZED CONVOLUTION ALGORITHM.

Abstract

Convolution and morphological operations are two fundamental classes of algorithms for computer vision. The authors develop a generalized convolution algorithm that processes the regular convolution, the binary morphological operation, and gray-level morphological operations on mesh-connected computers, using a snake sweeping mechanism to handle the interprocessor communication. This G-convolution algorithm is not restricted to any particular kernel size; in fact, it handles both large and small masks as well as even and odd masks. The algorithm is also not restricted to the one-pixel-per-processor case; instead, it handles images of any size on mesh array of any size. The image/processor mapping is only limited by the local memory of each processing element.