Development of new methods for codebook generation and optimization to enhance the vector quantization for still image compression

Abstract

Everyday, enormous amount of information is stored, processed, and transmitted digitally. This information requires high bandwidth and large volume of storage capacity. Designing an efficient compression technique is newlineessential to meet the recent growth of computer applications using images. newlineIn this thesis, efficient image compression techniques based on Vector Quantization (VQ) and Block Truncation Coding (BTC) are dealt with. VQ comprises of three stages: Codebook Generation, Image Encoding and Image newlineDecoding. The codebook generation phase plays a key role in vector newlinequantization. In this work, we propose new methods to i) generate and optimise newlinethe codebooks, ii) reduce the bit-rate required to store the compressed image and newlineiii) to improve the quality of the reconstructed images. Six new codebook newlinegeneration techniques are proposed in this work. In the first method, Simple newlineCode Generation (SCG), the codebook is generated in a simple manner. We then newlineimproved the SCG method and developed Ordered Codebook Generation (OCG) newlinemethod, in which the codebook is generated by sorting the training vectors. This newlineimproves the quality of the reconstructed images further. The third method is an newlineenhanced version of Pairwise Nearest Neighbour (PNN) method. In this, the newlinetraining vectors are sorted prior to the generation of codebook and the method is newlinenamed as Ordered PNN (OPNN). The time taken to identify the nearest pair of newlinevectors is reduced to a great extent (from several thousands to few hundreds of newlineseconds) and the quality of the reconstructed images is also improved. The OPNN method is further enhanced to get OPNNMM method newlinewhich reduces the time to a greater extent by merging multiple pairs of vectors newlinein one single iteration.Three novel ideas have been proposed to improve the quality of the reconstructed images. In the first method, the edge blocks are retained to avoid newlineragged edges in the reconstructed images. In the second method, the code vector.