Robust image compression for transmission over time-varying channels

We propose a rate-distortion (RD) approach to the optimization of image coding for transmission over time-varying channels. The information available about the channel error rate, is used to adapt the joint source-channel coding (JSCC) scheme. We focus on structured JSCC schemes to obtain realizations with feasible complexity. The image is hierarchically quantized, followed by unequal channel error protection matched to the importance of the bits. We allow the use of variable length codes in the source coder which provides adaptation to image statistics and higher compression performance. At the same time, we maintain error resilience by sending the synchronization information error free (through heavy protection). We show that efficient rate-distortion performance can be achieved by optimizing the overall system for the target rate and channel conditions. We present simulation results obtained on coding images for time-varying channels, which show that RD optimized JSCC schemes substantially outperforms conventional fixed length JSCC and can gain more than 6 dB in PSNR of the reconstructed frames.