| Abstract: |
All current state-of-the-art video error concealment schemes conceal the lost area through the reconstruction of 2-D patches. Reconstructed corrupted areas in the facial parts of head-and-shoulder video sequences, as in video conferencing applications, often suffer from objectionable artifacts. In this work, we present a novel video error concealment technique, which is assisted by Candide-3, a standard 3-D head-and-shoulder face model, for the reconstruction of corrupted facial regions with reduced artifact. The model is first adapted to facial images and then updated and tracked across frames, even in presence of lost macroblocks. The lost portions of the face are reconstructed through the projection of the adapted 3-D face model. The proposed concealment scheme has been experimented on sequences having facial areas such as Foreman, Carphone, News etc. and it outperforms some of the recently developed 2-D concealment schemes. © 2016, Springer Science+Business Media New York. |
