An adaptive leaky factor determination algorithm for AR-FGS

Leaky prediction is adopted by AR-FGS (FGS with Adaptive Reference) in the emerging standard H.264 SVC to improve the coding efficiency in low-delay applications. Leaky factor is critical to leaky prediction performance and the optimal leaky factor is close dependent on sequence content. However, no method has been proposed to determine the optimal factor. This paper proposes a flexible method that dynamically determines the optimal leaky factor for each frame through previous bit-rate of coded base layer. Combined with slightly modified bit-stream extraction procedure of JSVM, a better tradeoff between coding efficiency and drifting reduction is achieved, and the coding performance is further improved compared with using fixed leaky factor. Experiment results show that the proposed algorithm improves the coding efficiency over a wide range of bit-rates. Additionally, as a bit-rate control method, the modified extraction method has stable control on the bit-rate at frame level.