5/3 Motion Compensated Temporal Filtering (MCTF) is a tool for highly scalable video coding which has been recently studied by many researchers. This thesis presents several error concealment algorithms for 5/3 MCTF with lifting, which can be used to improve the quality of compressed video damaged by packet losses. In MCTF video, the low frequency subband frame, abbreviated as L-frame, contains most of the signal energy in any given Group-of-Pictures (GOP). We assume that one of these L-frames is lost. The proposed error concealment algorithms use the available data to reconstruct the missing L-frame. The simplest error concealment method considered in the thesis is Zero Motion Error Concealment. This method simply assumes zero motion through the damaged GOP, and averages the neighboring L-frames to reconstruct the missing L-frame. Another method called Motion Concatenation finds temporal pathways through the damaged GOP by connecting motion vectors available at the decoder, and copies the corresponding pixel values from the neighboring L-frames to the missing L-frame. Finally, Motion Re-estimation uses motion estimator at the decoder to find a motion vectors between two neighboring L-frames of the missing L-frame, and synthesizes the missing L-frame halfway between its neighboring L-frames. The overall error concealment system combines these three methods to maximize visual performance, as well as the Peak Signal-to-Noise-Ratio (PSNR).
展开▼
