The world is still trying and missing Deep Nostalgia, an AI technology provided by the online family tree service MyHeritage, which can add facial animation to still portrait photos. Almost every day, we encounter some small clips on the Internet in which the technology is used to animate still pictures, making the character smile, nod, wink and tilt the head. There is no doubt that this experience is wonderful. Although “Nostalgia” brings your photos to life, under the guidance of Dr. AN Rajagopalan, Chair Professor of Sterlite Technologies in the Electrical Engineering Department of IIT Madras, a new AI technology has been developed to help restore blurred or degraded images.
His image processing and computer vision laboratory at the Institute uses the function of artificial neural networks to restore images affected by rain streaks, raindrops, haze or motion blur. The research team found that it is not easy for a single neural network to identify the degraded part of the picture and clean it. Therefore, the team decided to divide the task into two phases. First, the network locates the degraded or fuzzy part. Then, in the next steps, the subsequent network uses the obtained information to restore the image.
“Our premise is to use the auxiliary task of degraded mask prediction to guide the recovery process. We prove that solving this auxiliary task will inject vital positioning capabilities in the network layer. We transfer this ability to the main through dedicated knowledge distillation In the restoration network, and use this additional knowledge to focus on the improvement of degraded areas.”
The method used by Dr. Rajagopalan’s team is described in detail in the IEEE titled “Method of Image Restoration Using the Degraded Perception Method of Knowledge Distillation”. The performance of this method is better than other previous strategies that attempt to restore old images. The team said it used a “public data set” of rain streaks, haze, raindrops and motion blur to test their model.
For the latest technical news and reviews, please follow Advertisement Shout on the following website Twitter, Facebook and Google News. For the latest videos on gadgets and technologies, please subscribe to our YouTube channel.
How to ignore mention notifications on WhatsApp