Abstract

In real-time video applications, error concealment is used to conceal the corrupted parts of the video sequence without retransmission. In this paper, an efficient multi-directional interpolation algorithm is presented for spatial error concealment. First, the significant edges of missing macroblock are only estimated using a directional edge analysis on the correctly received neighboring blocks. Second, an approximation for each missing pixel is computed along each significant edge. For two boundary pixels along each significant edge direction, the sum of magnitudes of gradients, which has the same quantized direction level as the edge direction, are computed on seven neighboring pixels. Finally, a weighted average of multiple approximations is computed using the sum of gradient magnitudes as weighting factors. Compared with the previous methods using multidirectional interpolation, experimental results show that the proposed method gives better reconstruction quality in terms of objective and subjective evaluations.