Video/Multimedia Technology

Software-based Enabling Technology for Video Services

A video signal, such as television, is a sequence of images or pictures. Each image contains a large volume of information just like the old saying "a picture is worth more than a thousand words". Consequently, a video signal has to be compressed or encoded so that it can be economically transported through a network or stored in a storage device. At the receiving end, it has to be decompressed or decoded for viewing. The techniques for encoding and decoding have to be standardized in order to avoid any chaos. There are two international organizations who deal with this issue. One is the International Standards Organization (ISO) Motion Picture Experts Group (MPEG) who recommends standards such as MPEG-2 and MPEG-4. The other is the International Telecommunications Union (ITU) who recommends standards such as H.261 and H.263 for videophone and videoconferencing applications. In the following, we will briefly describe two technologies related to MPEG-2 and MPEG-4 that are available for technology transfer.

To compress a video signal according to MPEG-2 or MPEG-4 standards requires a hugh amount of computation. It is often done by using specially designed hardware or chip sets. However, there are two major disadvantages about this approach. First, it is not flexible and difficult to make any changes, if needed. Second, the quality of the compressed video signal is somewhat fixed and difficult to improve further. We can eliminate these drawbacks by using a software-based approach. Since the required amount of computation is too excessive for a single processor to handle, we have developed an enabling technology for compressing video signals using distributed and parallel computing with multiple processors. On the other hand, a video compression standard only specifies a format (or syntax) for representing data inputs to the decoder and a set of rules (or a semantics) for interpreting them. This leads to two implications: First, there dose not exists a unified way of implementing the encoder, and second, continuing improvements in the encoder are possible. We have taken full advantages of the above implications by implementing our encoders using the software- based approach.

MPEG-2 is intended for high quality video compression. It has been used for Digital Video Broadcasting (DVB), High Definition Television (HDTV), Video-On-Demand (VOD), and Digital Versitle Disk (DVD) applications. We have used advanced techniques derived from our previous research for the software-based implementation of encoder as well as decoder. For a given degree of compression (or bit-rate) we can produce a better quality video than using a hardware approach, or for a given level of video quality, we can produce a video with less bit-rate. Our approach is also very cost effective.

MPEG-4 is a new standard currently under development. It is intended for content-based access, communication, and manipulation of digital audio/ visual objects which can be either natural or synthetic. In addition to high efficiency compression, MPEG-4 also considers error-resilient encoding in hostile environment such as wireless channels. MPEG-4 is also compatible with MPEG-2. Typical applications include wireless video- telephony, news gethering and editing, multimedia messaging, scientific visulization, games, etc. MPEG-4 is much more complex than MPEG-2. Hardware implementation is possible but only for a very specific application. Software-based implementation is ideal for various applications.

MPEG-4 Standards Participation

HKUST’s motion estimation research led to the contribution and the eventual acceptance of a fast motion estimation algorithm to the ISO/IEC 14496 (MPEG4) standard. It is the first technology developed in Hong Kong and China being accepted by the MPEG standard. HKUST’s algorithm was first submitted in Oct. 1998. After more than 18 months and 7 rounds of rigorous and stringent competitive evaluation by more than 300 experts worldwide, a modified version called Predictive Motion Vector Field Adaptive Search Technique (PMVFAST) was accepted into MPEG4 Part 7 in Mar. 2000.

With PMVFAST, a speed-up factor of up to 8000 with essentially no visual quality degradation is possible over the brute force full search motion estimation technique. A patent on PMVFAST and another fast search technique was filed in Mar 2000. Many companies have indicated interest in the techniques. In particular, we are now in the final stage of negotiation to transfer the technologies to a MPEG4 technology provider company.