 |
 |
 |
 |
 |
 |
||||
    | ||||
|
|
Seeing through the snowTechnology developed by Monash IT researchers is giving crime fighters the leading edge. RITA FELDMANN reports Most of us know it as 'snow' that annoying, flickering distortion that interrupts our view of an otherwise clear video or television picture. But for those trying to identify a particular individual or establish the exact sequence of events from a surveillance video recording, 'snow' or other sorts of video 'noise', can mean the difference between solving a crime, or not. Associate Professor Henry Wu, from Monash University's School of Computer Science and Software Engineering, leads a team which is among the world's best in turning unidentifiable video sequences captured by video surveillance systems into quality digital video images.
These techniques remove noise, the umbrella term for sources of visual distortion, such as snow, that can corrupt a recorded video. Noise in recorded video can be caused by a number of things: thermal and electronic noise in a CCD (charge- coupled device) camera, fluctuations in lighting, transmission noise, signal inter-ference from other electronic equip- ment and low-performance recorders or video-editing equipment. How this noise appears visually depends on the source of the distortion. The team found success recently with a poor quality video central to a criminal investigation. The Victorian Forensic Science Centre provided a tape of an incident outside a nightclub that was captured by an external security camera. The tape revealed what the police were looking for: the escape vehicle. However, due to the speed of the car, the intensity of light in the foreground and variables associated with the recording camera, the car they wanted to identify appeared only as five or six blown-out lights in the background, says Dr Wu. "When they approached us, initially we weren't totally confident that we could produce the results they wanted," he says. "We knew that the filters and associated techniques we had devised were the best in the world for clearing images corrupted by particular types of noise known as 'salt and pepper' and Gaussian noise," he explains. "However, the real-world problem of the noisy surveillance video was more complicated. We weren't sure that the noise on the video could be modelled by the types of noise we had designed the filters for." As Dr Wu explains, the first thing the team did was digitise the analog video recorded on the VHS video tape. Using advanced digital video equipment, they converted the original video on the analog VHS tape into an uncompressed digital video format, known as D1 format. This process doubled the vertical resolution of the video material and minimised the possibility of introducing further noise into the video. After analog to digital conversion, the digital video sequence was generally clearer. However, the vehicle in the recording could still only be identified as a cluster of lights. The next step was to enhance the sequence of images by increasing the spatial and temporal resolution. This was achieved using the group's own 'de-interlacing' filter. When applied to the getaway car tape, the results Dr Wu obtained were remarkable. With the light-noise removed and with the application of twice as much spatial and temporal resolution, the six blown-out lights gave way to an image of a car, with a very distinctive set of tail lights. Finally, the police had the evidence they needed to confirm the make of the car involved in the incident. At present, the image processing software developed by Dr Wu's team is in its prototype version. However, the filters have the potential for widespread application. Dr Wu says there is also great potential for use of the software in medical diagnostics and satellite imaging. "X-ray images may be contaminated either in the imaging process or digitisation process, especially if the film has aged or been scratched over a period of time," says Dr Wu. "Our image processing techniques can help to eliminate or reduce the noise and restore the image for record-keeping by medical practitioners." With expressions of interest in the research coming from the US and China, Dr Wu says for now the group is continuing its work on improving the design of de-noising filters for colour image processing. This work is being carried out with the assistance of a one-year Australian Research Council Discovery Project grant. For more information on the services offered by the team, or for collaborative research links, contact Associate Professor Henry Wu on +61 3 9905 3255 or visit www.csse.monash.edu.au/~hrw/ |
|
In technical-speak, it is known as digital video image processing and enhancement and involves cleaning up and improving the
resolution of video sequences using filters special algorithms run on digital computer.