Video Error Concealment Techniques for Multi-Broadcast Reception of Digital TV

The transmission of digital TV signals to mobile receivers is often error-prone. As most TV broadcasting techniques provide only moderate error robustness, horizontal lines of consecutive image blocks are lost during decoding of the received video signals. In order to ensure high viewing experiences, these lost slices have to be filled by error concealment techniques. However, the reconstruction qualities of classical approaches which exploit spatial, temporal, or spatio-temporal signal correlations are not convincing yet. In the future, mobile TV receivers will support different broadcasting techniques in parallel. As a result, an erroneous high-resolution video signal and a correctly received low-resolution video signal, both representing the same TV service, will often be available. Focusing on the outlined scenario for multi-broadcast reception of digital TV signals, this thesis introduces the novel category of inter-sequence error concealment algorithms. The basic idea is to fill lost slices of the high-resolution video signal by the interpolated low-resolution video signal. Since the images of this reference signal are often cropped and delayed, robust spatio-temporal image alignment is crucial. By including a pixel-based or a feature-based alignment scheme, the proposed concealment algorithms provide excellent visual qualities and outstanding reconstruction qualities of up to 41 dB PSNR. Classical concealment techniques are outperformed by up to 15 dB PSNR. To further enhance the reconstruction quality, several extensions are introduced. First, the alignment robustness and the interpolation quality are increased. Subsequently, a classical temporal approach is incorporated as an alternative concealment mode to cope with low image qualities of the reference signal. Novel aspects include robust mode selection, enhanced motion estimation, and the reconstruction of the displaced frame differences from the reference signal. As a last extension, spatial refinement tackles blurring of concealed image blocks. Missing spectral components are recovered in a frequency selective way based on approximation and extrapolation principles. By combining all relevant extensions, the PSNR gain adds up to 20 dB with respect to classical concealment. Finally, inter-sequence error concealment is adapted to multi-broadcast reception of two erroneous high-resolution video signals. While spatial alignment can be omitted, classical concealment of blocks, being lost in both video signals, and drift compensation in predictively-coded frames are novel aspects. Again, high visual qualities are obtained and classical concealment is outperformed by up to 15 dB PSNR.