Wavelet Based Lattice Vector Quantization Scheme for Multiple Description System

Recent years have witnessed the increase in demand for fast, efficient, and high quality multimedia communication applications over the wired and wireless networks. The constraint in bandwidth has paved the way for researchers to embark their efforts in finding efficient compression techniques for image and video. The demand for high quality of data communication has open up the research area in multiple description coding to flourish. Data transmission via network communication system with a single channel is always likely to fail because of noise and interference which lower the system performance. In order to increase the robustness and therefore offering high performance, multiple descriptions coding (MDC) has been proposed. MDC produces several separate descriptions, while decoding individual description can reproduce the source with certain accuracy. MDC related works incorporate compression schemes especially vector quantization (VQ) and lattice VQ techniques. In this thesis, various A2 LVQ quantizing techniques are introduced and their proofs are simulated in personal computer as well as using mathematical derivations. A new algorithm for multistage A2 Lattice Vector Quantization (MA2LVQ) is introduced. The hybrid LVQ quantizers (Hy-MLVQ) techniques are also presented. The new schemes use the combination of two lattice types. A unique method to calculate and construct the A2 lattice structure is presented using the triangle number and square numbers. In this development, mathematical and geometrical expressions are used to provide better understanding of the construction process. Then wavelet based image coding using the various LVQ techniques are presented. In this work, the MA2LVQ is used to encode the wavelet transformed image or subbands with the help of the proposed adaptive normalization technique where significant improvement in performance is achieved. Wavelet based image coding also utilizes the new hybrid multiple bitplane-MA2LVQ (HBM-LVQ). The HBM-LVQ utilizes three side decoders. The scheme also proposes a perfect estimation decoder and operates at very low bit rate. The third part of this thesis is the hardware implementation of the ordinary A2 and low complexity A2 LVQ hardware architectures are proposed. Then, the MA2LVQ hardware architecture is developed using several A2 LVQ quantizers that are arranged consecutively. The performance of synthesizable MA2LVQ is conducted using RTL and gate level simulations. Finally, the MA2LVQ is implemented using the Cyclone II FPGA and displays the output using the VGA controller in LCD monitor. Experimental results show that the proposed MA2LVQ and Hy-MLVQ quantizers allow reduction of quantization distortion due to the increment of lattice point density in the subsequent sublattice or quantizer. The adaptive normalization technique for wavelet based image coding using MA2LVQ quantizer also offers significant improvement in performance. Moreover, the HBM-LVQ can reach very low bit rate (below 0.2 bpp) and exhibits good performance in all side decoders. The synthesizable low complexity A2 LVQ architecture performs well up to 35.97 nanoseconds (ns). Finally, the implementation of MA2LVQ using low complexity A2 reduces the use of logic and register elements up to 47% as compared to the use of the ordinary A2 LVQ architecture.