The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical Engineering University of Southern California

Technical Report USC-SIPI-375

“Robust and Flexible H.264 Video Streaming Techniques”

by Xiaosong Zhou

August 2006

Digital video streaming is finding a much broader range of applications nowadays. The quality of these applications is challenged by the inevitable transmission errors in most existing communication channels. At the same time, transmitted video stream may need to be adjusted due to changes in network conditions or user interactions. The objective of this research to design robust and flexible video coding and transmission mechanisms in order to improve the error resilience performance of compressed video stream, and to enable flexible bit stream adjustments. We propose two different error resilient video coding and transmission schemes for H.264 video streaming applications. The first method aims to eliminate or largely reduce error propagation for off-line H.264 encoders based on a concept called "alternative macroblock coding (AMC)". The AMC scheme utilizes alternative coding and bit stream replacement to stop error propagation. Two different methods to create alternative predictions are proposed to encode the alternative macroblocks efficiently. The experimental results show that the proposed method is able to largely reduce error propagation without introducing large bit rate overhead. Another more general error resilient video coding scheme utilizing multi-path predictive coding is proposed. Two different methods to calculate the expected decoder distortion are also proposed in order to evaluate the error resilience performance of different predictions. An adaptive reference frame selection scheme is designed to incorporate this measurement into the mode decision process. It is shown by experimental results that the encoded stream has a much better error resilience performance by using the proposed scheme without large degradation of the compression efficiency. We propose the use of difference pictures to compensate mismatches introduced by the bit stream adjustments so as to preserve the quality level of the received video. The difference pictures can be encoded efficiently so that they will not introduce large bit rate overhead. Experimental results show that the proposed methods outperform existing ones significantly in terms of preserving the compression efficiency of compressed stream. The flexibility of video streaming applications can be greatly enhanced by using the proposed methods.

This report is not currently available in PDF format for downloading. Contact the Signal and Image Processing Institute for information on its availability.