IIT-M professor creates AI-based system to restore damaged photographs
MERIN JAMES
Do you have photographs that are faded or been damaged by moisture, and you don’t know how to get them restored? Dr AN Rajagopalan is working on a solution. A professor in the Electrical Engineering department at IIT Madras, he has come up with a new technology to restore photographs. “Photographs are timeless - I have seen many people wondering what to do with some of the photographs that have suffered distortions. It is a massive challenge for many. I have created an artificial neural network to restore pictures that have been degraded by raindrops, haze or even motion blur. It is a nonlinear mapping method — for eg, certain parts will be more degraded than the other; so first the neural network will identify and isolate that portion. Our premise is to use the auxiliary task of degradation mask prediction to guide the restoration process. We demonstrate that solving this auxiliary task injects crucial localising ability in network layers. We transfer this ability to the main restoration network using attentive knowledge-distillation and focus on the refinement of degraded regions by exploiting this additional knowledge,” says Dr Rajagopalan.
He tells us that the eventual goal is to make this process accessible to all. “There is a huge potential but the main bottleneck is that it requires very high memory power. The mobile phones that are presently available in the market don’t have that much memory storage. We have to wait and see how we can incorporate this,” he adds.