Integrated fast implementation of mathematical morphology operations in image processing

Abstract

A systolic architecture for two basic mathematical morphology operations is presented. Mathematical morphology has been proven to be a very useful tool for applications such as image restoration, image skeletonization, pattern recognition, machine vision, etc. The advantages of the design stem from the fact that it has a pipeline period alpha =1, requires simple communications, and exploits the simplicity of the morphological operations to make it possible to implement them in a linear target machine without using multiprojection, although the starting algorithm is a generalized two-dimensional convolution. The optimal partitioning strategy for a constrained mapping when the number of PEs is fixed is given, achieving 100% processor utilization, and a systolic chip design which is capable of real-time video performance is proposed as the basic processing element. >