Image Coding and Video Source Models base on Finite-State Machines and Wavelet Transforms

• Main Authors: Chen Chien-Fa, 陳建發
• Other Authors: Wu Chien-Hsing
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/20662810051298272684

碩士 === 國立雲林科技大學 === 電子與資訊工程技術研究所 === 86 === Vector quantization (VQ) is an efficient method for lossy data compression,By integrating VQ with a finite-state machine, we can construct a finite-state VQ(FSVQ) capable of reducing bit rates and speeding up the codingprocess. FSVQ has been widely used in applications such as speech,image and video processing. On the other hand, the discrete wavelet transform(DWT) is a relatively new signal processing technique that providesan elegant way to multi-resolution analysis of signal sources. In this thesis, we combine the techniques of DWT and FSVQ to work out a powerful method for image coding. We conduct extensive experiments to examine the achievable compression ratio and PSNR by appropriately allocating the number of bits for each of the muli-resolution codebooks. Our experimental results indicate that thecombined DWT/FSVQ techniqueoutperformsthe discrete cosine transform (DCT) based image coding method which exhibits prominent block effects at low PSNR. When applied to a 256-gray level image of size 256 by 256(pixels). the DWT/FSVQ technique achieves a compression ratio of 40 (0.2 bpp) at PSNR=30dB. We also develop a set of Markov models for representing the traffic generationprocess of an MPEG video source by characterizing the embedded stochasticfinite-state machine. An effective state-space decomposition algorithm isproposed to provide a baseline tool fo0 selecting a finite-statemachine withan irreducible state transition diagram.