Spatial Error Concealment Based on Bezier Curves

This paper presents a geometric error concealment scheme for DCT-based image/video data based on the Bezier polynomials (BP). The proposed scheme makes use of the average edge direction and local curvature (extracted from healthy blocks around the damaged block) as boundary conditions to create an n-degree BP. The resulting BP is then used to interpolate the edges of the lost block, as well as to directionally reconstruct its low frequency data. When multiple edges are identified crossing the lost area, a cost function defined in terms of the local and global edge curvatures is used to find the best match for each missing edge. Experimental results show that our approach provides near perfect reconstruction and excellent subjective quality of the restored data, outperforming current linear interpolation schemes in the literature