Realtime convolution and morphology on VLSI-based mesh: a systolic approach

An outline is presented of a modular design of an enhanced mesh network for image processing and vision applications. Algorithms for implementing convolution and morphology on this network are studied with time analysis. In contrast with earlier work in this area, in which it is assumed that the image data is readily loaded into the system, the global broadcasting is addressed in this study. The authors exploit the common operations among correlation, convolution, and morphology to design a special-purpose VLSI mesh that is capable of performing template mapping, filtering, dilation and erosion at a very high speed for real-time applications such as robotics or autonomous vehicles