Concealment aware mode selection for power-rate-distortion optimized H.264/AVC encoder

In our recent work, an analytic power-channe/error-rate-distortion (PERD) model was developed to estimate the end-to-end distortion of power constrained portable devices in real-time video communications over bandwidth limited and lossy channels. A random mode selection scheme is implemented after obtaining optimal percentages of intra, inter and skip mode through a constrained optimization process. In this paper, advanced solution to the problem of mode selection for each macroblock (MB) is proposed. A concealment aware PERD model is first derived, which takes error concealment at the decoder into consideration to estimate the end-to-end distortion. By using this model, optimal percentages of intra, inter and skip modes that minimize decoder distortion are obtained under any power and rate constraints. An intelligent mode selection scheme is then generalized by ranking the priority of MB according to how well it gets reconstructed if lost. Experimental results show that, the new framework consistently achieves better power-rate-distortion (P-R-D) performance with little extra computation load, which is critical in real-time communications of portable devices.