Resilient Digital Video Transmission over Wireless Channels using Pixel-Level Artefact Detection Mechanisms

Recent advances in communications and video coding technology have brought multimedia communications into everyday life, where a variety of services and applications are being integrated within different devices such that multimedia content is provided everywhere and on any device. H.264/AVC provides a major advance on preceding video coding standards obtaining as much as twice the coding efficiency over these standards (Richardson I.E.G., 2003, Wiegand T. & Sullivan G.J., 2007). Furthermore, this new codec inserts video related information within network abstraction layer units (NALUs), which facilitates the transmission of H.264/AVC coded sequences over a variety of network environments (Stockhammer, T. & Hannuksela M.M., 2005) making it applicable for a broad range of applications such as TV broadcasting, mobile TV, video-on-demand, digital media storage, high definition TV, multimedia streaming and conversational applications. Real-time wireless conversational and broadcast applications are particularly challenging as, in general, reliable delivery cannot be guaranteed (Stockhammer, T. & Hannuksela M.M., 2005). The H.264/AVC standard specifies several error resilient strategies to minimise the effect of transmission errors on the perceptual quality of the reconstructed video sequences. However, these methods assume a packet-loss scenario where the receiver discards and conceals all the video information contained within a corrupted NALU packet. This implies that the error resilient methods adopted by the standard operate at a lower bound since not all the information contained within a corrupted NALU packet is un-utilizable (Stockhammer, T. et al., 2003). Decoding partially damaged bitstreams, where only corrupted MBs are concealed, may be advantageous over the standard approach. However, visually distorted regions which are not accurately detected by the syntax analysis of the decoder generally cause severe reduction in quality experienced by the end-user. This chapter investigates the application of pixel-level artefact detection mechanisms which can be employed to detect the visually impaired regions to be concealed. It further shows that heuristic thresholds are not applicable for these scenarios. On the other hand, applying machine learning methods such as Support Vector Machines (SVMs) can significantly increase the decoder’s capability of detecting visual distorted regions. Simulation results will show that the SVMs manage to detect 94.6% of the visually impaired MBs resulting in Peak Signal-to-Noise (PSNR) gains of up to 10.59 dB on a