A study of error -resilient interleaving with applications in the transmission of compressed images and video
Abstract
Three interleaving algorithms are developed and evaluated to improve the quality of compressed images or video delivered over error-prone channels. The algorithms operate on macroblocks (MB), the multiple coding entities, and the transform coefficients, respectively. Two of the interleaving algorithms focus on maintaining the synchronization of video coding units composed of variable length codes (VLC) in a bit reversal error environment. The third algorithm redistributes spatially correlated transform coefficients of images to temporally separated transport units to be delivered in a burst error environment. The first interleaving algorithm can specify the start position of each MB in a compressed video bitstream. It extends error resilient entropy coding methods by utilizing reverse direction interleaving at the end of each interleaving slot. The algorithm is implemented in the form of transcoders placed before and after the channel for an MPEG-4 Simple Profile bitstream. A simple, syntax-based codeword repair method is also proposed so that the transcoder generates an MPEG-4 compliant bitstream which can then be decoded with a standard MPEG-4 decoder. The second interleaving algorithm referred to as the nested interleaving algorithm is developed in order to provide three levels of synchronization in a compressed video bitstream: the MB, the discrete cosine transform (DCT) block, and the VLC levels. This algorithm is also described in the form of transcoders. Since three-level interleaving provides synchronization on the VLC scale, bit errors can be detected in a VLC unit based on syntax conformity. The detected errors can be repaired syntactically so that the transcoder generates an MPEG-4 compliant bitstream. To reduce the effect of burst errors or lost packets, a DCT coefficient interleaving method is developed for JPEG. Since the interleaving disrupts the correlation between coefficients, the entropy coder that follows will generate longer code words. The relationship of the DCT interleaving with the incurred redundancy is investigated using a pseudo-rate analysis framework known as ρ-domain analysis. It is shown that the optimal interleaving map can be obtained only by exhaustive search. To restore the lost coefficients due to burst error, a maximum a posteriori (MAP) estimator for the lost coefficients is derived in closed-form by converting the Gauss-Markov Random Field a priori probability of pixels into the probability density function of the DCT coefficients.
Degree
Ph.D.
Advisors
Delp, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.