Project aims to create smarter video DCU centre is researching specialised techniques to search and edit video footage

Wish you could flop into a chair at home after a long day's work and instruct your television set to show you only the news items or programmes that show Roy Keane - or, for that matter, Catherine Zeta Jones, Bertie Ahern, or Homer Simpson? Or maybe you'd prefer to have highlights from the week's GAA matches automatically edited into a compact video?

Or to take a film such as Thelma and Louise and have your video player edit it into a neat package showing only the scenes that feature Brad Pitt? If research at Dublin City University's Centre for Digital Video Processing comes to commercial fruition, you may be able to do just that. The "centre" - which is actually spread across a couple of beautiful new buildings on the constantly-evolving DCU campus - is a cross-disciplinary research centre and a collaboration between the School of Computer Applications and the School of Electronic Engineering.

The centre, directed by Prof Alan Smeaton, has six faculty members, along with four post-graduates and 14 doctoral students. Funding and support come from a number of outside sources, including the National Software Directorate, Enterprise Ireland's Informatics 2000 and ATRP2002 programmes, the EU Marie Curie Fellowship Programme, the Irish Research Council for Science, Engineering and Technology, Dublin City University, RINCE, Sun Microsystems, Parthus and Logica.

The centre defines its mission as researching and developing techniques and tools to automatically analyse and index digital video information and allow content-based operations such as browsing, searching, alerting, filtering and summarisation.

Confused? Never fear; Prof Smeaton is not one for idle chit-chat when there are plenty of research projects to be demonstrated. A visitor to the facilities is whisked off to peer over student shoulders at a succession of computer screens.

First up is a comprehensive set of projects that fall under the title Físchlár, Irish for "video programme". According to Prof Smeaton, most of the centre's projects are folded into Físchlár, so that the centre can demonstrate a coherent working system to users. About 2,000 people on campus use Físchlár, not just people involved with the centre, but a wide range of students and academics, with one segment of the programme specifically targeted at DCU's nursing programme.

"We have a philosophy that if we do anything good with video, we put it into Físchlár," notes Prof Smeaton. Físchlár collects television broadcasts from eight terrestrial channels, around the clock teletext broadcasts, educational videos, and a lot of research work.

Físchlár is then subdivided into sections. 'Físchlár TV' captures terrestrial broadcasts as well as teletext, and works in some ways like the commercial American product TiVo. It can capture, index, and replay television broadcasts - but with pizazz. For example, the system can look for frames it believes to be important from the way in which images are displayed in a frame, and choose 200 to 400 of them as content summaries. It whittles those down to 30 frames, to form a mini "key frame summary" of a programme.

The frames can then be presented back to an end-user in a variety of ways, each of which has particular strengths depending on what a viewer is searching for. "The idea is you can find a section of a programme quickly, whereas with traditional film editors you have to wind the film forward and backward," Prof Smeaton says.

Another interesting application is Físchlár News, which captures the nightly broadcast of RTÉ1 news along with its teletext. Two years of news have been recorded, creating a valuable archive that can be searched by text (using the teletext matched to its screen image).

Thus, says Prof Seaton, if you want just the stories featuring Tony Blair and George W. Bush, you can probably retrieve most of them using a text search, then create links between stories, and cross-index.

At the far end of experimentation 'Físchlár Trek', which adds real intelligence to video indexing capabilities. "Físchlár Trek builds in a face detector and group recognition and analyses whether an image is indoors or outdoors, in the city or country. On the audio side, it can detect whether music is in the background, it can read text that appears within a frame, and tell whether a single person or group of people is speaking. But this is a system where we have lots of work to do." The centre is working to enable these video applications to be accessible by a handheld device such as the iPaq computer. Indeed, making video mobile is the central interest of a commercial spin-off of the centre, a company called Aliope Ltd, which had staff at the recent 3GSM exhibit in Cannes to flog their wares.

"Aliope takes some of the centre's work and brings it to market," explains Mr Fergus Keenan, engineering manager. Aliope offers tools for producing and publishing video and other content to mobile devices, and for creating mobile portals.

"This is bridging the content producer and the mobile operator," he says, showing how a user can capture a bit of video on a computer - in this case, a football player scoring a goal - give it a caption, format it for mobile distribution, and send it out to a device.

Aliope hopes to capitalise on the developing market for MMS messaging, which allows users to send images and video and audio clips to mobiles.

Over in another building, Dr Noel Murphy's researchers and students in electrical engineering do "the analysis that allows Alan to do the things he's been demonstrating", says Dr Murphy, a lecturer at the centre. He has six groups working in various areas.

Lecturer Dr Noel O'Connor explains some of the projects. "This lab produces the next generation of tools that eventually will be plugged into Físchlár," he says. In one demo, a computer is set the task of analysing exactly where "the exciting parts" of a sports match are, using clues such as the rise in crowd noise. "Eventually, this could produce a five-minute summary of a match and send it to a mobile."

Another project tries to teach a computer to detect specific objects rather than an event. The example here is the figure of Bart Simpson: his distinctive shape isn't too difficult for the program to spot., but set it a more ambiguous shape and the computer's slow analysis demonstrates this is a project in its early stages.

"The work of the centre is really to try to bridge the gap between the implementation and the concepts," says Dr O'Connor. "We want to find ways to make content that is understandable by humans, understandable to computers."