An investigation of digital image compression techniques

Abstract

By its nature, digital image data contains redundant information which, if removed, can reduce the amount of bandwidth required to transmit it, or reduce the space required to store it. In this thesis the characteristics of image data that give rise to redundancy, and one technique that can be used to reduce it, are discussed. The bandwidth/storage reduction technique considered is transform-based image data compression.
The underlying theory of the transform-based technique is described and implementation issues are addressed. Of particular importance for implementation of the technique is the use of an efficient entropy coding method. In this context,
Hufihian and arithmetic coding are two methods of reducing statistical redundancy; their underlying principles are reviewed and described in this thesis.
One of the disadvantages of transform-based techniques is the inherent and considerable amount of computation required. This thesis describes research performed to detennine the suitability of using a multi-processor system architecture
to reduce the time required to perform compression operations. It is shown that use of such a system, implemented using a Transputer network, does reduce the processing time. Furthermore, it is shown that when more processors are added to the network the time required for compression reduces, but it is also shown the compression achieved is reduced.
Results are presented that quantify the amount of compression achieved, the amount of distortion introduced into the reconstructed image and the speed of operation when systems consisting of up to eight processors are used.