What could be more frustrating than looking at a memorable moment of your life captured by a camera in poor definition? How do you feel when watching your favorite childhood TV program recorded on a VHS cassette on your recent 4K TV monitor?
Super Resolution can be a solution for the above but also a large number of other applications ranging from astronomy, to microscopy and satellite imaging. The idea behind Super Resolution is to synthetize missing pixels in a video frame using existing color, shape, structure and semantic information available. The simplest solutions use interpolation through linear filtering, by creating new pixels using their neighbors. However, they do not perform that well in challenging regions of an image, such as when interpolation the texture of the fur of a cat. More sophisticated Super Resolution techniques based on artificial intelligence (AI) have been proposed as a remedy to the latter weaknesses.
In AdMiRe project, Super Resolution will be applied as a tool to match the quality of professionally produced content in studio environment. AdMiRe relies on consumer grade capture devices such as webcam and smart phone, laptops and tablet cameras to capture remote participants and to integrate them in a studio environment. In a typical professional editing, the resolution of video varies between HDTV and 4K. This is above the typical resolution in consumer devices, especially when considering that cropping could be applied to the captured video. Nobody wants to appear of blurry on the national TV. Super Resolution come to the rescue here by increasing the definition of the captured video in addition to its contrast, brightness and color correction. Super Resolution is a challenging task because its performance depends on a large number of parameters such as the lighting conditions, the type of the device used, the distance of the person from the camera and the amount of up-sampling desired. In addition, other challenges such as complexity, power consumption, latency and real-time processing should be considered!
Jonathan Aeschimann / Touradj Ebrahimi – EPFL