Rate-distortion optimized reference picture selection

Most common video coding standards employ spatiotemporal prediction techniques to increase coding efficiency. As a result, compressed video data is sensitive to the transmission error due to error propagation. An adaptive reference picture selection (RPS) is proposed to improve the error resilience in real-time video transmission. A global rate-distortion optimization (RDO) is developed to intelligently select the reference picture from a reference picture list, such that the end-to-end distortion is minimized. The coupling variables in the global RDO are decoupled by applying primal decomposition approach iteratively. Experimental results in H.264/AVC indicate that the proposed adaptive RPS significantly outperforms the ACK and NACK mode of the classic RPS, and error resilient RDO in H.264/AVC reference software as well.