Research On Error Resilience Of Video Encoding And Transcoding

Due to increasing demands of multimedia applications and the limited transmission bandwidth, the video coding technique is critical and important. However, unreliable transmissions can degrade the integrity of the bitstream, which leads to poor video quality. As a result, more attentions have been received on the research of video error resilience. It is always a challenge for the researchers that, the intensity of error-resilient video encoding can be adjusted adaptively based on the transmission condition and video contents. Different video applications have their own standards, resolutions, frame rates, and transmission condition, so video transcoding is also crucial for the video applications. Similarly, transcoded bitstreams are transmitted in unreliable routines, so researches on the video transcoding's error resilience techniques also become necessary. As a result, it is of theoretical significance and practical value to conduct an in-depth research on the error resilience of video encoding and transcoding.In chapter 1, the significance of the research work is presented together with a brief summary of the present research status. Then, the main research contents and the chapter structure of the thesis are introduced.In chapter 2, an error-resilient video encoding scheme based on joint intra and reference selection refreshes is proposed to meet the need of transmission with noticeably fluctuating status. Compared to traditional schemes that often employ a single and fixed tool, the proposed scheme enables two refresh tools to function in suitable scenarios. The video contents are first analyzed, and every macroblock's expected distortion is calculated. In order to improve the region-of-interest (ROI) quality, an ROI extraction approach is employed, and the ROI macroblocks' distortions are weighted. Then, based on the transmission condition, bitrate, and expected distortions, the refresh number, refresh mode, and optimal referencing distance are computed.In chapter 3, an error-resilient video encoding scheme based on H.264 flexible macroblock ordering (FMO) mechanism is proposed, which not only meets the need of prioritized transmission environment, but also takes advantage of FMO's inherent error resilience function. A motion area extraction method is first utilized, and the motion area's error sensitivity is then computed according to both transmission condition and input pictures'features. Based on the error sensitivity, the optimal FMO encoding mode is selected, which makes the error-resilient intensity more adaptive. To lower the complexity, most information for computations is from the encoding domain; the product of the encoding process.In chapter 4, since all current schemes take the encoding quality as the single transcoding aim, the research on the error resilience of frame skipping video transcoding is conducted. A frame skipping video transcoding method based on the adaptation of both encoding quality and error resilience is proposed in this chapter, which enables the dynamic balance between the coding efficiency and the error-resilient capability from the perspective of transcoding architecture. The sliding window mechanism for frame skipping transcoding is first used to ensure accurate frame rate, controllable delay, and small frame rate fluctuation. Then, both quality impact factor and error impact factor are evaluated, based on H.264 compressed domain. Finally, according to impact factors and the transmission condition, the optimal frame skipping allocation is determined.In chapter 5, an error-resilient video transcoding method based on H.264 redundant slice technique is proposed. This method improves the error resilience through adaptively embedding ROI redundant slices into the original bitstream, and the transcoded bitstream is flexible in terms of error-resilient intensity adjusting. The original bitstream is first decoded, and its ROI part is extracted. Then, distortions of embedding and not embedding redundant slices are calculated as well as bits consumptions. Based on these calculations, the ROI part's rate-distortion performance is calculated. Whether the redundant slice should be embedded is determined according to this performance. After dependency analysis and ROI enlargement, the redundant slice is produced by the entropy coding and embedded into the bitstream when needed, which doesn't involve the second-round encoding. The final chapter concludes the new achievements of the whole research and the prospect of the future research.