Error resilience and concealment techniques for packet video transmission

Since the quality of compressed video is vulnerable to errors, video transmission over unreliable Internet is very challenging today. In this thesis, two important issues about robust packet video transmission are investigated.;The first issue is error resilient (ER) video compression. Motivated by two-hypothesis motion-compensated prediction (THMCP), we first propose an error resilient video coding technique, where two-hypothesis and one-hypothesis predictions are alternately used to encode the video stream. We use some schemes to determine which kind of prediction should be used, so that in some cases of frame losses, the propagated error can be first decreased to some extent before it spreads to the subsequent frames. Then we extend the previous work for THMCP to a more generalized one, i.e. the reference frames can be some distance from the current frame, instead of only the immediately preceding ones. Three types of prediction patterns are proposed and implemented by the generalized B pictures in H.264/AVC. In the case of a single frame loss during the transmission, the induced error propagation is analyzed and the closed-form expression for error energy is derived.;The second issue is error concealment (EC) for packet losses in video transmission. We first propose an error concealment algorithm to reconstruct a lost INTER-frame in the odd/even temporal sub-sampling MDC. The neighboring frames in the error-free stream are used to temporally interpolate the lost frame, based the preserved motion vector in this correct stream. To further improve the reconstructed video quality after the lost one, we also propose a multi-hypothesis error concealment (MHC) algorithm. In addition to error-concealing the lost frame, MHC also applies temporal interpolation to some additional frames after the frame loss so as to reduce propagated error quickly. After discussing the EC algorithms for INTER-frame losses, we propose two algorithms to error-conceal a lost INTRA-frame. The novelty is that not only the INTRA-frame but also the subsequent INTER-frames are error concealed based on the strong correlation between adjacent pixel values. In addition, motion compensation is used to reconstruct the INTER-pixel which has an INTRA-pixel in its motion trajectory. Finally, we propose an edge-directed error concealment algorithm to recover lost slices in FMO-encoded video. The strong edges in a corrupted frame are estimated first, which are then used to direct the recovery of the structure regions and the remaining erroneous regions.