Implementation of Soft Computing Approaches on Vector Quantization Based Clustering For Image Compression

Abstract

This thesis mainly contributes in the area of image compression by the application of soft computing (SC) techniques to ensure high compression ratio with better reconstructed image quality. An attempt has been made in the present study for the application of various SC techniques to improve the reconstructed image quality of multimedia devices. In this regard, improved SC techniques and hybridization of these SC techniques as well as new clustering are proposed for performance enhancement of multimedia devices. Over the years, many techniques have been suggested for image compression. These techniques include vector quantization, Multi-level image thresholding and fractal image compression. Some disadvantages of these techniques for image compression are: (i) design of suitable codebook, threshold and affine transforms are crucial and improper design causes performance degradation, (ii) convergence time is increasing exponentially with size of the design (iii) final outcome purely depends on the initial random design (iv) attain a considerable high compression ratio causes degradation in reconstructed image quality. In traditional techniques, problem formulation must satisfy mathematical restrictions and may suffer from numerical problems. Further, in a complex design consisting of a number of elements, the optimization of several elements using the traditional techniques is a very complicated process and sometimes gets struck at local minima resulting improper design. SC techniques provide a global optimal solution or nearly so. Among the SC techniques, modern heuristic techniques such as bat algorithm (BA), cuckoo search (CS), hybrid CS, teaching learning based optimization (TLBO), differential evolution (DE) gravitational search algorithm (GSA) and pattern search (PS) algorithms are used for optimization of codebook and threshold. An attempt has been made in the present study for the application of various AI techniques to improve the reconstructed image quality of the technique.