A new object-based variational approach for MPEG2 data recovery over lossy packet networks

In this paper, we propose an object-based variational approach for the decoding of MJPEG and MPEG1-2 video sequences. This new method improves the visual quality of the reconstructed sequence by reducing the impact of cell losses, due to the transmission of the sequences over lossy packet networks. Generally, two kinds of approaches are suggested to tackle this problem : a preventive "network" approach tends to provide a more robust bit-stream in order to help the recovery of the lost data, for instance by adding redundancy to the bitstream, and a concealment "image" approach, that tends to reduce visual impact resulting from the packet loss. The proposed method combines "network" information provided by the transmission protocol, with an adapted "image" decoding algorithm. Several experimental results demonstrate the efficiency of the decoding method : lost areas are largely recovered, compared to a standard decoding, both on the background and on the objects.