Efficient implementation of motion-compensated video compression and its application in multimedia communications

Abstract

Video compression plays an important role in various multimedia applications. With the rapid deployment of the Internet, more and more distributed multimedia applications, which not only include voice and data but also image and motion video, have been emerging in recent few years. However, most of the video services currently provided on the Internet such as Web TV and video surveillance, are broadcast in nature. In other words, they are a one-way communications link, transmitting video data from servers to users. Conversational video services such as video telephony and video conferencing are still not popular because of the high cost of a video encoder bore at the user side. Obviously, a low-cost software-based video codec is one of the feasible solutions for supporting conversational multimedia communications on the Internet. This thesis is focused on several algorithmic issues relating to the realization of a real-time software-based motion-compensated video codec and its applications in multimedia communications. In this thesis, we first develop an efficient block matching algorithm to reduce the high computational loading of motion-compensated video compression. A new adaptive search algorithm is first formulated to adaptively select the search pattern for the block matching algorithm in order to locate the motion vector according to different local characteristics of the test block. A new motion tracking algorithm is then derived to locate the search origin for the fast search algorithm by exploiting the high motion correlation in both the temporal and spatial domain. Combining the motion tracking technique with the new adaptive search, a very efficient motion estimation algorithm is formulated. Based on the proposed fast block matching algorithms as well as several algorithmic enhancements on DCT coding and quantization, a real-time software-based H.263 video codec is then developed. In our implementation, a stack-based solution is also derived to support multiple decoding channels as required by various multimedia applications. Based on the H.263 video codec, we have implemented a software-based desktop video conferencing system according to the ITU-T H.323 Recommendations. The system can support up to 6 people making possible a conference within a local area network. The frame rate achieved is about 5-8 frames/s and 15-20 frames/s for CIF and QCIF sequences respectively.