Document Type

Thesis

Date of Award

1-31-1993

Degree Name

Master of Science in Electrical Engineering - (M.S.)

Department

Electrical Engineering

First Advisor

Sotirios Ziavras

Second Advisor

John D. Carpinelli

Third Advisor

Dionissios Karvelas

Abstract

Image processing is used for manipulation of pictorial images. Image analysis applications are typically characterized by the need to process large quantities of image data. Some of the important transformations or operations which are car¬ried out by image processing systems are translation, scaling, superposition and rotation. Algorithms have been developed to carry out these transformations on image regions represented by quadtrees. Gargantini introduced an algorithm to translate an image region represented by a linear quadtree or leafcodes. A linear quadtree is a space efficient data structure used for storing digital images. Ziavras et.al. have proposed a modification of Gargantini's algorithm which makes it much more efficient. Ziavras's algorithm translates as many leaves as possible without splitting them. This thesis carries out a comparative analysis that involves these two algorithms. The comparison is based on results obtained from simulation of these algorithms for a hypercube parallel computing system. Simulation results are obtained for a single pixel and multiple pixels per processing element (PE) of a hypercube parallel computing system. In the case where multiple pixels are stored in each PE, a binary image of size 2P x 2P is subdivided into quadrants of equal size and then stored in an n-dimensional hypercube. It is shown that Ziavras's algorithm performs much better than Gargantini's algorithm when p is larger than n. Gargantini's algorithm may perform better when a single pixel is assigned to each PE.

Share

COinS