Abstract
In recent years, the demand for real-time transmission of video over networks has greatly increased. However, due to the unreliability of packet-switched networks, streaming video has encountered the problem of packet loss. To solve this problem, multiple description coding was proposed, which adds redundancy between packets, and we may estimate the original video data when receiving part of packets. However, since video coding employs motion compensation, there will be error propagation in the decoding if some packets of previous frames are lost. To solve this problem (which is also called “drift”), we propose a new algorithm based on multiple descriptions and partial/leaky predictions.First, we introduce the concept of base/enhancement parts of MDSQ, which is referred to as Base MDSQ. That is, we can extract the reconstruction of the base part from either description. Then a new video coder combined with Base MDSQ and two-layer partial/leaky prediction is proposed to solve the problem of drift. Since the loss probabilities of the base/enhancement parts are different, we assign different leaky factors for each of them.From the simulation results, we show that the proposed Base MDSQ coder with leaky prediction outperforms the normal MDSQ coder with leaky prediction by about 0.09dB in most sequences. However, Base MDSQ with partial prediction has the similar performance as normal MDSQ with leaky prediction. Base MDSQ with leaky prediction outperforms Base MDSQ with partial prediction at the expense of high complexity.