X-Tree based Unequal Loss Protected packetized coding

摘要

A novel Multiple Description Coding method, named Packeted Multiple Description Coding (PMDC) is presented in this paper. First, we separate the wavelet transformed image into several sub-images and code each sub-image by a method called Unequal Loss Protected Multiple Description Subband Coding (ULPMDC). With this algorithm, we can first form two sub-descriptions, which contain packets from the image, and then we transmit them by two different channels. If no more than one sub-description per sub-image is lost, the proposed algorithm will get very good reconstruction quality. If more sub-descriptions per sub-image are lost, the reconstruction quality is severely degraded. To provide better quality of the reconstructed image, we then modify the packetizing process, and propose a novel packetized MDC framework, named Cross Packetized Multiple Description Coding (CPMDC). In this method, instead of applying ULPMDC on sub-images, we first code the whole image into two descriptions. Then we split each description horizontally and vertically to form row packets and column packets, respectively. The simulation results show that if only some row packets or column packets are lost, the reconstruction quality is still as good as the former scheme. If we lose both columns and rows, CPMDC will still be better than PTSQ.