Error concealment for packet video

Abstract

Two new algorithms are proposed for error concealment in the presence of cell/packet loss for packet video. Under packet switching network environments like Ethernet or ATM, some video packets may be lost or discarded due to transmission delay or channel congestion. Because video is normally compressed, small data losses may completely destroy the received video signal. Thus, lost block recovery or concealment becomes an essential issue for video communication over networks. The proposed algorithms use block DCT coding with one pixel overlap. Without any assumption on the smoothness of block boundaries, the combination of these two algorithms provide consistently good performance in lost block recovery. Experimental results show that the quality of the recovered video is much better than that obtained with maximally smooth methods. One additional advantage of the one pixel overlap structure is that it removes blocking effects automatically. Blocking effects are annoying problems for all block transform based image or video CODECs. With simple averaging operations on the overlapped pixels, all blocking effects can be suppressed. One-pixel overlapped block coding generally decreases the compression ratio for a given quantization step size. However, increases in PSNR achieved by overlapped block coding largely compensate for this effect. Experimental results show that for a given PSNR, the decrease in compression ratio is negligible. To demonstrate the proposed algorithms, a standard CCITT H.261 software encoder and decoder were created using C++. With minor modification, the standard CODEC is easily combined with our proposed error concealment algorithms to provide high performance block loss recovery for packet video.