BBC R&D are involved in the research and development of the next generation of video compression technologies. Since the advent of digital television, video compression has been central to making and distributing television programmes. The sheer number of pixels in digital television pictures - up to 52 million per second for HDTV - makes it imperative to use compression in all parts of the broadcast chain. This means in cameras, in production networks, in archives, as well as for broadcasting and streaming to the viewer.
Central to the BBC's work over the next few years will be the development of a new MPEG standard, codenamed High-performance Video Coding (HVC). HVC is targeted at higher resolutions - HDTV and above - to provide the best compression for very high resolution services, made possible by the latest generations in display technology. The aim of the standard will be to obtain a doubling of compression efficiency over the current state of the art - represented by the MPEG AVC (a.k.a. ITU-T H264) standard - for these resolutions.
At the same time as display resolutions are increasing in the home, the development of internet streaming services such as YouTube and iPlayer is increasing very rapidly, as is the use of mobile devices for viewing video. These applications require video to be decoded in software, and may use a wide variety of resolutions. So it is important that the new standard is also capable of operating with low complexity, and of trading off complexity against compression performance in a very flexible way. Ideally the new standard will be capable of providing performance and complexity gains at the same time in some configurations.
These requirements have been captured in the MPEG Call for Proposals for the new standard, and will guide the development of the technology over the coming years.
So how does compression work, and what will be new in the new standard? A compression system consists of three basic stages: prediction, approximation and entropy coding. Entropy coding is the final stage of compression, where symbols which describe the video are turned into a bit stream. This is essentially a problem of producing codes for the symbols that relate to the probabilities of their occurrence. An example of an entropy code is Morse code, which uses short symbols for very common letters and longer ones for rare letters. This process can be addressed mathematically, and near perfect solutions have been engineered and incorporated into previous standards.
So most of the likely improvements will come through finding better ways to predict video, and better ways to approximate what is left after prediction. In video compression the most powerful predictions are temporal, making use of the fact that one frame of video is quite close to others captured before and after. This means that only the differences between frames need to be considered. Tools to capture these differences have got increasingly sophisticated through generations of video standards, and there is still a lot of mileage in making these predictions more sophisticated still.
In addition to clever prediction techniques, a key theme in the new standard is likely to be better ways to approximate video, using everything that we have learnt about the human visual system. In this respect, video compression lags behind audio compression, which has used perceptual coding for many years. The visual system is more complicated, but increasingly tools are being developed, such as texture synthesis, which can approximate what should be there using very compact sets of parameters.
Can we reach the target of 50% reduction? It will be tough, but we believe we can, and through our involvement with HVC we will be at the cutting edge of development of these technologies.
Click Back to top
~RS~a~RS~rd~RS~q~RS~~RS~z~RS~42~RS~){kind=small}
