New spatial error concealment with texture modeling and adaptive directional recovery

A spatial error concealment technique based on the sequence-aligned texture modeling and the adaptive directional recovery is proposed in this work. Texture modeling is achieved using the sequence alignment technique. It captures the local variation and the global trend of image textures in corrupted regions with surrounding uncorrupted pixels, and provides the best texture model under a given cost function. Results of texture approximation are in a form of aligned pixel pairs, and the desired local texture model can be obtained by connecting them with line segments. With the derived texture model, geometric interpolation is then used to recover lost pixels adaptively based on pixel locations. There are four candidate pixel sequences to recover lost pixels, and one of them is selected for the concealment purpose. Since a pixel sequence that provides good estimation of lost pixels should also offer a good estimate of pixels surrounding the corrupted region, the minimum-mean-squared error (MMSE) between original and estimated uncorrupted pixels is used as the selection criterion. Extensive experimental results are given to demonstrate that the proposed error concealment technique outperforms several benchmark methods in both objective and subjective tests.