|
Video codecs operating at integral multiples of 64 kbps are well-known in visual communications technology as p * 64 systems (p equals 1 to 24). Originally developed as a class of ITU standards, these codecs have served as core technology for videoconferencing, and they have also influenced the MPEG standards for addressable video. Video compression in the above systems is provided by motion compensation followed by discrete cosine transform -- quantization of the residual signal. Notwithstanding the promise of higher bit rates in emerging generations of networks and storage devices, there is a continuing need for facile audiovisual communications over voice band and wireless modems. Consequently, video compression at bit rates lower than 64 kbps is a widely-sought capability. In particular, video codecs operating at rates in the neighborhood of 64, 32, 16, and 8 kbps seem to have great practical value, being matched respectively to the transmission capacities of basic rate ISDN (64 kbps), and voiceband modems that represent high (32 kbps), medium (16 kbps) and low- end (8 kbps) grades in current modem technology. The purpose of this talk is to describe the state of video technology at these transmission rates, without getting too literal about the specific speeds mentioned above. In other words, we expect codecs designed for non- submultiples of 64 kbps, such as 56 kbps or 19.2 kbps, as well as for sub-multiples of 64 kbps, depending on varying constraints on modem rate and the transmission rate needed for the voice-coding part of the audiovisual communications link. The MPEG-4 video standards process is a natural platform on which to examine current capabilities in sub-ISDN rate video coding, and we shall draw appropriately from this process in describing video codec performance. Inherent in this summary is a reinforcement of motion compensation and DCT as viable building blocks of video compression systems, although there is a need for improving signal quality even in the very best of these systems. In a related part of our talk, we discuss the role of preprocessing and postprocessing subsystems which serve to enhance the performance of an otherwise standard codec. Examples of these (sometimes proprietary) subsystems are automatic face-tracking prior to the coding of a head-and-shoulders scene, and adaptive postfiltering after conventional decoding, to reduce generic classes of artifacts in low bit rate video. The talk concludes with a summary of technology targets and research directions. We discuss targets in terms of four fundamental parameters of coder performance: quality, bit rate, delay and complexity; and we emphasize the need for measuring and maximizing the composite quality of the audiovisual signal. In discussing research directions, we examine progress and opportunities in two fundamental approaches for bit rate reduction: removal of statistical redundancy and reduction of perceptual irrelevancy; we speculate on the value of techniques such as analysis-by-synthesis that have proved to be quite valuable in speech coding, and we examine the prospect of integrating speech and image processing for developing next-generation technology for audiovisual communications.
|
