Efficient Image Computations on Reconfigurable Meshes

The reconfigurable mesh combines two attractive features of massively parallel architectures, namely, constant diameter and a dynamically reconfigurable bus system. Being mesh-based, the reconfigurable mesh is eminently suitable for applications in image processing and computer vision. Common wisdom has it that buses are entities created in support of communicational needs of the task at hand. The authors have recently proposed a new way of looking at a reconfigurable bus system. They have shown that buses can be used as computational devices as well as topological descriptors. The purpose of this work is to present additional results of this on-going project. Specifically, we show novel algorithms for image segmentation and labeling, along with a number of algorithms for describing the shape of objects in the image. We discuss using the buses to compute the area, perimeter, and convex hull of every object in the image.