Constant quality constrained bit allocation for leaky prediction based FGS video streaming

Leaky prediction based FGS (Fine Granularity Scalability) can achieve better coding efficiency than the baseline FGS. However, for leaky prediction based FGS (L-FGS), constant quality constrained bit allocation, i.e., how to optimally allocate bits given the current channel bandwidth, is still an open problem. In this paper, based on the accurate R-D (Rate-Distortion) model developed in our previous work, we propose a constant quality constrained bit allocation scheme for L-FGS. The proposed scheme is a combination of offline and online processes. During the offline stage, we perform the L-FGS encoding and collect the necessary feature information. At the online stage, given the transmission bandwidth at that time, we quickly estimate the R-D curves of a sequence of consecutive video frames based on our previously developed R-D model and then perform the corresponding bit allocation using a sliding window technique. Experimental results show that our proposed bit allocation algorithm can achieve much more smooth video quality than the traditional uniform bit allocation under both CBR (constant bit rate) and VBR (variable bit rate) channels.