Communications Research Centre Canada
www.crc.gc.ca
Common menu bar links
A New CRC Video Codec Outperforms Current Video Compression Technology
| As anyone involved in the electronic entertainment business knows, the viewers of today are both sophisticated and discerning. They demand high-resolution images, brilliant colours, and seamless life-like movement. To top it all off, they want content delivered where they want it, when they want it - whether that be to the rec-room HDTV, the office computer, the smart phone or hand-held gaming device. "People want more content and more choices," says André Vincent, Manager of CRC's Advanced Video Systems group, "but at the same time, they want faster delivery and better image quality." These new mobile technologies, he explains, paired with an increasing appetite for high-definition and 3-D video products, is putting enormous pressure on companies in the business of delivering content. Raw video requires an enormous amount of data, and the higher the resolution, the higher the data rate required to achieve the flawless image and movement that viewers demand. The higher the data rate, the more bandwidth it takes, and bandwidth is limited and expensive. "A lot of program material is delivered by satellite before going to cable or out to your wireless device. If you can reduce the data rate, this reduces the cost of transmission." The answer has been to develop codecs: systems that compress video for transmission then decompress it at the viewer end. The most commonly used system is based on the MPEG standard introduced in 1993. It has continuously been refined and optimized through to the latest version, MPEG-AVC/H.264-4. Like all compression technology, the MPEG standard uses a set of mathematical rules to analyze, then transform the raw video data into a more condensed code. In the case of MPEG, the data is transformed using a discrete cosine transformation (DCT). The compressed data that is eventually transmitted is a set of coefficients based on this mathematical transformation. The information is then decoded at the other end to recreate the image data. "But," stresses Vincent, "there is no magic. Compression comes at a cost and the cost is some reduction in picture quality." And, he says, while the discrete cosine transformation that the MPEG standard is based upon has served us to date, it may not be enough to provide a further 50% reduction in bit rate as expected for the next generation of video compression standard. |
(a) Original picture, (b) compressed at 0.2 bit per pixel with CRC-WVC technology, and (c) compressed using MPEG-4/AVC at 0.2 bit per pixel. The quality of the picture compressed using CRC's technology is 1.3dB better than that compressed using MPEG-4/AVC. |
Vincent's lab decided to do the latter, using directional wavelets as the foundation for their new prototype codec, CRC-WVC.
The mathematical "transform" that one uses to analyze and code the data, explains senior scientist Demin Wang - the CRC-WVC codec developer - acts something like a filter. Each transform has its own particular properties: it picks up certain characteristics of the image while allowing other information to pass through. The trick to choosing the best and most efficient transform for video compression is to find one that picks up and codes for the information most important to viewer experience, while leaving the irrelevant bits behind.
Using a transform based on directional wavelets, says Wang, does exactly that. The traditional DCT transform used by MPEG performs a two-dimensional transform on small blocks of the image, coding changes in the horizontal and vertical axes extremely well, but performing less well on information off the horizontal or vertical axis. A transform using directional wavelets, however, is highly sensitive to unpredictability - attributes like edges moving at odd angles or curled or wavy lines - precisely the kind of information viewers are glued to as James Bond hurtles over a scaffold or the claws of a monster lash out at your character in Final Fantasy 11.
"We don't need to spend a lot of bits, or data, on uniform areas," explains Wang. "We just need to concentrate on important features. That's the benefit of using directional wavelets. Images that contain a lot of lines or contours are coded very precisely."
While still in the early stages of development, viewer tests are showing that the new CRC-WVC performs as well as, or better than, MPEG-4, especially on certain kinds of program material. And CRC-WVC compression efficiency, says Vincent, is especially good on high definition video sequences. In addition to video compression, the lab will also be investigating other applications for the technology in medical imaging, fingerprint analysis, and super HDTV. It can be applied, stresses Vincent, to any application where precise edge information is important.
For more information contact André Vincent, Manager, Advanced Video Systems at andre.vincent@crc.gc.ca or 613-998-2299.
