Jun-Horng Chen
Error Concealment, Matrix Completion, SVD, Nuclear Norm
This work aims to improve the quality of spatial error concealment by utilizing the low-rank characteristic of natural image blocks. In contrast with the conventional approaches which usually exploit the correlation between neighbor image blocks and boundary pixels to predict the corrupted pixels, this work regards the error concealment as a matrix completion problem. The proposed approach preserves the local texture information by minimizing the nuclear norm, and thus recovers the corrupted image blocks. Furthermore, this work also proposes the edge discontinuity across the boundary pixels should be diminished in the completed pixel matrix. The simulation results will show that the proposed approach indeed improves the quality of error concealment. Compared with the non-normative spatial error concealment of H.264/AVC, the improving gain is about 2 dB of PSNR.
Important Links:
Go Back
