A low-complexity error concealment algorithm for video transmission based on non-local means denoising

In video communication, the concealment of transmission errors is important to mitigate error propagation and allow for a pleasant visual quality. The previous spatio-temporal error concealment algorithm, known as Denoised Temporal Extrapolation Refinement (DTER) used a spatial denoising algorithm to reduce the imperfectness of the temporal extrapolation. However, the computational complexity of DTER is extremely high. In this paper, a fast algorithm based on Non-Local Means (NLM) is proposed for video transmission. The proposed fast strategy mainly includes two steps. First, the search window can be narrowed down by using the characteristic that the similarity of any two image blocks will decrease with increasing Euclidean distance. Afterwards, a successive elimination method is used for cutting out some redundant points. By making use of these two steps, our proposed algorithm achieves 10~30 times faster than DTER without sacrifice in picture quality. Moreover, simulation results clearly show the proposed fast algorithm outperform other existing error concealment algorithms.