3-D model assisted facial error concealment technique using regenerative particle filter based tracking

This paper proposes a novel approach for error concealment in corrupted frames, using the Candide 3.0 face model. In head and shoulder video sequences, for example in live chats and video conferencing, error concealment performance may be improved using methods that are specifically designed to handle the structural details of the human face. The proposed approach uses perspective projection to adapt the model onto the face in the video and then performs tracking using regenerative particle filters. The pixel information from past frames is transformed into the current (corrupted) frame through this model. The simulation results are promising.