Nonuniform Threshold Trellis Coded Quantization for Image Transmission Through Noisy Channels

ZH Sun,CW Chen

DOI: https://doi.org/10.1109/iscas.1997.621963

1997-01-01

Abstract:We propose a joint source/channel coding scheme to transmit image through binary noisy channels based on 2-D discrete cosine transform (DCT) and nonuniform threshold trellis coded quantization (NUTTCQ). After the discrete cosine transform, the source image is decomposed into several subsources according to the transform coefficient positions, i.e. the same frequency coefficients in different DCT blocks are grouped together as a single subsource. The subsources are modeled by a generalized Gaussian distribution (GGD), and the statistical parameters of the GGDs and channel characteristics are used to construct the scalar codebooks for NUTTCQ at both the encoder and the decoder. The transform coefficients are quantized by these fixed-rate quantizers and transmitted through noisy channels. The steepest descent method and iterative schemes are employed to determine the optimum bit allocation among the subsources subject to the constraints of the average coding rate and allowable maximum bits to each sample. Neighborhood relationship is employed to limit the searching space when a bit is to be allocated to a certain subsource. Simulation results show that the performance is very promising