Research

Video Networking Laboratory

The Video Networking Laboratory is the UK's longest established visual Lab., originating in 1969 and internationally known for its innovative video related research.

Professor Ghanbari, Head of the Lab, has summarised our research goals and capabilities in this set of slides.

A summary of our recently completed EPSRC project (Fuzzy-Logic Transcoded Video Stream Controller) can be found here.  We were also specially commissioned by the Office of Communications (Ofcom) to report on Future Performance of Video Codecs.

Video Coding

The core of our work is in video compression, with various aspects of compression tools for motion estimation, entropy coding, scalability and temporal filtering first appearing within the laboratory. The design and implementation of video codecs is a speciality from very low bit rate (less than 20 Kbit/s) to HDTV ( about 20 Mbit/s) and Super HDTV (30-155 Mbit/s). A best-selling book on codecs, published by the IEE describes these innovations, and a further title describes advanced codecs from the perspective of the laboratory's activities. The laboratory and its staff have devised numerous coding techniques which have ultimately been incorporated into video coding standards. The most notable of these is scalability for which Prof Mohammed Ghanbari, who leads the lab, became a Fellow of IEEE. Sprite coding in MPEG-4 also represents a significant contribution and we are amongst the pioneers of spatial transforms for motion estimation. Based upon our reputation, we are predicting the future performance of video codecs in the next two decades for the U.K. Government's Office of Communications (Ofcom).  We have prepared various examples of our work, including video clips.

Video Transcoding

A principal branch of our activity is video transcoding, by which an already compressed bit-stream can be converted to different formats without the need to decode. This has numerous applications such as: video distribution to heterogeneous decoders and rate control to avoid network congestion. Several methods of video transcoding have been pioneered by us and are universally known in the research community. Most of our publications have appeared in high quality journals such as the IEEE Transactions series and well-known research databases record a consistently high citation record. Many of these achievements were and continue to be accomplished through projects funded by the EPSRC and BT Exact, with the final report of EPSRC funded projects consistently rated as 'Internationally Leading' or 'Outstanding'.  There are some small examples of transcoding available.

IPTV

Transmission of video over unreliable networks such as the Internet to obtain acceptable end-to-end quality of service is another major contribution of ours. This now exists as IPTV and by monitoring and measuring of network quality of service in conjunction with video transcoders, network utilisation is maximised, while maintaining an optimal quality of service. Existing methods of congestion control are unsuitable for video streaming as they produce an erratic quality at the receiver. All analysts agree that IPTV's importance is set to soar in the next five years. As a last resort to improve video quality, the effect of dropped packets by one of our numerous methods of concealing lost information to improve video quality . Currently we are investigating the use of fuzzy logic for congestion control under an EPSRC research grant.  You can see a simple example of how congestion control can radically improve the quality of streamed video.

Video Quality Metrics

Another activity for which we have received international recognition is the development of an objective quality model for video services. It has long been argued that Peak-Signal-to-Noise-Ratio (PSNR) is not a realistic indicator of video quality. Over two EU projects; MOSAIC & TAPESTRIES and involvement in the Video Quality Experts Group (VQEG) activity, we have devised objective video quality measures. Our most notable contribution to date is the extraction of quality measures directly from the compressed bit-stream, without the need to decode it. The method uses the difficult approach of making no reference to the source video and has the practical application of monitoring video quality as the video stream crosses the network. This undoubtedly improves network intelligence and opens up a new way of resource allocation, while maintaining quality of service. Our methods of measuring objective video quality measures correlate well with the associated subjective test results, for which we have examples.

Image/Video segmentation

Segmentation of images is another important part of our activity. The correct identification of objects in still images and video has important implications for the success of many emerging compression and analysis tools such as: object based coding in MPEG-4; database search engines like MPEG-7; and the authoring tools of MPEG-21. In our work, we exploit the saliency between regions inside a binary partition tree, leading to a method whereby tentative relationships between the salient and semantic are described. Our research is producing some powerful segmentation tools that can be tailored according to the complexity of application. The MELISA project successfully applies our methods to recognise and track sportsmen in events, such as horse racing and football. The methodology separates the image processing from the analysis by making image retrieval, tracking and segmentation a database rather than image-processing problem. The work continues to unearth interesting findings and is feeding nicely into image and video retrieval, by which under MPEG-7, segmented objects can be described by shape, colour, motion and texture for automatic or manual browsing.

In the next example, each incoming frame is first automatically segmented by our tree-based segmentation algorithm and yellow team players (indicated by green squares) are detected through region space analysis and retrieval. One of them (indicated by a red cross) is tracked by using our TUSE tracking algorithm. The main advantage of the system is that it does not involve any camera related operations; the camera is not required to be calibrated and can be either stationary or moving.

Animated sequence showing player tracking with the TUSE algorithm.
(Double click the play arrow to start)

Video watermarking

Video watermarking or the embedding/hiding of information inside images themselves is vital when distributing pictures over networks. Without it, ownership of a particular document cannot be established. In our investigations, we study the impact of watermarking on video compression and ways in which a signature can be embedded in a sequence of video frames, with the minimal level of visibility and maximum resilience to attack.

Video over wireless

In line with current trends, we are also active in engineering the transport of video over wireless networks such as Bluetooth, WiFi and WiMax. Complex interactions between packetization, radio frequency noise, cross-traffic congestion and desired video quality arise, as our recent publications reveal. One of the major applications for low bit-rate video is the broadcast of video to mobile handheld devices, with the emergent DVB-H standard being a focus of such activities. Video transport must now be achieved across heterogeneous networks which employ a variety of fixed and wireless network types, with differing bandwidths and protocols. The Internet or converged IP networks will deliver video before it crosses a wireless access network to the display device, whether that wireless interconnect is short range like Bluetooth or longer range like DVB-H. Delivery of best video quality within this heterogeneous environment is the main thrust of this work in progress. We have just finished an EPSRC grant which investigated error control for broadcast video linked to hierarchical modulation and again the final report was rated Leading to Outstanding or Internationally Leading. There are some examples available, including video clips.

As a by-product of our investigations, Rouzbeh Razavi has created an IEEE 802.11 simulator for Matlab, with a Java front-end.  A download is available.  The normal usage conditions of acknowledgement apply and the release is made in good faith but without any obligations.