Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Computer Science

First Advisor

Sitharama Iyengar


There has been tremendous research interest in the areas of computer and robotic vision. Scene understanding and parallel image processing are important paradigms in computer vision. New techniques are presented to solve some of the problems in these paradigms. Automatic interpretation of features in a natural scene is the focus of the first part of the dissertation. The proposed interpretation technique consists of a context dependent feature labeling algorithm using non linear probabilistic relaxation, and an expert system. Traditionally, the output of the labeling is analyzed, and then recognized by a high level interpreter. In this new approach, the knowledge about the scene is utilized to resolve the inconsistencies introduced by the labeling algorithm. A feature labeling system based on this hybrid technique is designed and developed. The labeling system plays a vital role in the development of an automatic image interpretation system for oceanographic satellite images. An extensive study on the existing interpretation techniques has been made in the related areas such as remote sensing, medical diagnosis, astronomy, and oceanography and has shown that our hybrid approach is unique and powerful. The second part of the dissertation presents the results in the area of parallel image processing. A new approach for parallelizing vision tasks in the low and intermediate levels is introduced. The technique utilizes schemes to embed the inherent data or computational structure, used to solve the problem, into parallel architectures such as hypercubes. The important characteristic of the technique is that the adjacent pixels in the image are mapped to nodes that are at a constant distance in the hypercube. Using the technique, parallel algorithms for neighbor-finding and digital distances are developed. A parallel hypercube sorting algorithm is obtained as an illustration of the technique. The research in developing these embedding algorithms has paved the way for efficient reconfiguration algorithms for hypercube architectures.