Switching error concealment algorithm based on optimal decisions for performance and complexity

Abstract

The proposed paper discusses two switching error concealment algorithms: SECA (Switching Error Concealment Algorithm) and OSECA (Optimal Switching Error Concealment Algorithm) to efficiently switch among three existing error concealment methods for optimal performance and complexity tradeoff. SECA uses the motion vector statistics in the neighborhood of the lost MB (Macro Block) to make the switching decision. SECA is better than the best-PSNR method by 0.14 dB in PSNR and better than the best-win-rate method by 1.49 % in win rate, with 63 % time complexity reduction. OSECA further considers the correlation between surrounding pixels of the lost MB and the pixels in the reference frame. An optimization problem is formulated to derive the optimal decision thresholds. For OSECA, the PSNR performance is further improved by 0.24 dB from SECA and the time complexity reduction from the existing work can increase to 69 %. On different testing sets of the lost MBs, the proposed SECA and OSECA perform consistently against existing methods. Compared to SECA in the testing sets, OSECA has larger PSNR improvement (by up to 0.23 dB) against the best-PSNR method, and has larger time complexity reduction (by up to 71 %) against existing method.