Parsing robustness in High Efficiency Video Coding - analysis and improvement

One problem in the current High Efficiency Video Coding design is that the entropy decoding, or the parsing process of one frame may depend on syntax elements on other frames. Thus, a small error may lead to immediate failure of decoding of the current frame and all the following inter frames. This paper analyzes the cause of this vulnerability of HEVC bitstreams. Based on the analysis, we further present two solutions to make a tradeoff between the coding efficiency and the error robustness by cutting down the error propagation caused by temporal motion vector prediction. One solution is to redesign some code tables so that the dependency on other frames' syntax elements can be broken. The other solution is to insert recovery points for the coded bitstream, so that the decoding process can recover from errors. The pros and cons of these two solutions are discussed. Experimental results show that the proposed methods can solve the parsing problem in HEVC with a marginal influence on the coding performance and provide significant improvements for the decoded video when there are errors.