Advances in Recursive Per-Pixel End-to-End Distortion Estimation for Robust Video Coding in H.264/AVC

This paper is focused on expanding the applicability of the recursive optimal per-pixel estimate (ROPE) of end-to-end distortion, to the H.264/AVC standard. One open question involves the complexity of cross-correlation terms that appear in the case of subpixel prediction and other pixel filtering operations. Several efficient model-based solutions are proposed. Another open question involves the largely ignored effects of rounding operations, whose cumulative impact may seriously degrade the estimate accuracy. Two approaches are proposed for rounding error compensation: one appeals to the maximum entropy principle, while a low complexity alternative is based on quantization theoretic approximations. The former effectively estimates the distribution of the decoder reconstruction, thereby significantly broadening the applicability of ROPE to all additive distortion measures. Simulation results for H.264/AVC with 1/4-pel prediction show that the proposed ROPE extensions achieve fairly high estimation accuracy, while maintaining low complexity. Another set of results demonstrates the level of overall coding gains achievable by exploiting such improved end-to-end distortion estimation.