A Novel Hybrid Open-Close Loop FGS Coding Framework Based on Key Reference Picture Selection

This paper proposes a novel fine granularity scalability coding framework with hybrid open-close loop structures (KRPS-FGS) based on key reference picture. The open-structure with single prediction loop is exploited to code non-key frames to achieve the best coding efficiency, which uses the highest quality image as the reference picture for both base and enhancement layer. To control drift, some key frames are periodically inserted, which exploit close-structure with AR-FGS to achieve the trade-off between coding efficiency and robustness. Key frames are only predicted from key frames, and no enhancement information is introduced into the prediction loop of base layer, so the drift is confined between two adjacent key frames. To optimally allocate bit-rate among enhancement layer, an optimized bit-stream extraction method is proposed, which gives more priority to the enhancement layer of key frames as compared with that of non-key frames, because the contribution of key frames to the quality of whole sequence is greater than that of non-key frames. Simulation results show that the proposed framework outperforms AR-FGS about 1~2 dB over a very wide rang of bit rate. Only at the very low bit-rate points close to base layer bit-rate, the performance decreases slightly.