Gaussian mixture-based generative approach for H.265/HEVC error concealment

Loss of data in video transmission over an error-prone channel is inevitable. Video error concealment is a powerful tool for resolving the effects of these errors. In this paper, we conducted several experiments with various video sequences to estimate the distribution of motion vectors (MVs) surrounding the corrupted area. As a result of the experiments, the proposed method suggests an innovative generative clustering approach using the Gaussian mixture model (GMM). The proposed algorithm first measures the GMM’s parameters based on the available surrounding MVs. Then each MV is assigned to exactly one cluster. Next, each cluster’s likelihood is calculated, and the one is chosen based on the maximum likelihood criteria. Finally, new MVs are generated for the chosen cluster, and the MV that can minimize the boundary distortion is selected for the corrupted block. Comparison concerning recent state-of-the-art techniques shows progress in PSNR and SSIM for H.265/HEVC. The proposed method improves the average PSNR by up to 5.67 dB and an average increase of 0.1135 in SSIM. Moreover, the computational complexity of the proposed algorithm is context-adaptive and shows better performance for the videos with relatively uniform motions over the missing areas.