Date of Award
Doctor of Philosophy (PhD)
James L. Schwing
Typical image processing and computer vision tasks found in industrial, medical, and military applications require real-time solutions. These requirements have motivated the design of many parallel architectures and algorithms. Recently, a new architecture called the reconfigurable mesh has been proposed. This thesis addresses a number of problems in image processing and computer vision on reconfigurable meshes.
We first show that a number of low-level descriptors of a digitized image such as the perimeter, area, histogram and median row can be reduced to computing the sum of all the integers in a matrix, which in turn can be reduced to computing the prefix sums of a binary sequence and the prefix sums of an integer sequence. We then propose a new computational paradigm for reconfigurable meshes, that is, identifying an entity by a bus and performing computations on the bus to obtain properties of the entity. Using the new paradigm, we solve a number of mid-level vision tasks including the Hough transform and component labeling. Finally, a VLSI-optimal constant time algorithm for computing the convex hull of a set of planar points is presented based on a VLSI-optimal constant time sorting algorithm.
As by-products, two basic data movement techniques, computing the prefix sums of a binary sequence and computing the prefix maxima of a sequence of real numbers, and a VLSI-optimal constant time sorting algorithm have been developed. These by-products are interesting in their own right. In addition, they can be exploited to obtain efficient algorithms for a number of computational problems.
"High Performance Issues in Image Processing and Computer Vision"
(1992). Doctor of Philosophy (PhD), Dissertation, Computer Science, Old Dominion University, DOI: 10.25777/c9w6-sk31