Trade-off and applications of source-controlled channel decoding to still images

Abstract

Summary form only given, as follows. For image transmission, using a new channel decoder, we present improvements in image quality leading to a much more graceful degradation in case of degrading channel conditions. The APRI-SOVA based on the Viterbi algorithm exploits the residual redundancy and correlation in the source bit stream without changing the transmitter. For three different quantizers (applied after a discrete wavelet transform), we show and discuss the trade-off between increasing source coding performance in case of no channel errors (uniform threshold (UT)-generalized Gaussian (GG)-pyramid vector quantizer (PVQ)) and the decreasing improvement using the APRI-SOVA in case of equal error protection (EEP) for noisy channels (PVQ-GG UT). We develop a means to judge the applicability of the APRI-SOVA by considering the remaining correlation of the coded bits (much for the simple UT, little for the complex PVQ), together with a semi-analytical way to calculate the expected improvement. Simulation results for EEP and additive white Gaussian noise show improvements for the LENNA image of up to 1.8 dB (UT), 1.3 dB (GG) in PSNR and no gain for the PVQ, with UT outperforming the other quantizers and thus providing gains of up to 4 dB in PSNR and up to 0.75 dB in E/sub S//N/sub 0/ when choosing the right quantizer. Even greater gains of up to 2.2 dB (UT) in PSNR and 0.5 dB in E/sub S//N/sub 0/ can be encountered when applying combined source and channel coding together with unequal error protection (UEP).