Computer Science

As society has moved from SD to HD to 4k, and now looks to move from 4k to 8k, the need to remaster and bring older media up to date is becoming more and more necessary. Some high-resolution televisions even come with real-time upscaling technology built in, but these models are limited by the lightweight hardware that comes with the TV. I intend to make high quality, real-time upscaling one of my research goals. Smaller, faster versions of the state-of-the-art could be fine-tuned for greater performance in consumer deployment by combining multiple models for specialist tasks. However, studio remastering scenarios may not always benefit from increased resolution, and rarely do they benefit from real-time speed. In these cases, the state-of-the-art should be advanced for maximum effectiveness. One recent advancement is using a combination of low-resolution and high-resolution images, across a range of time to synthesise high-resolution images at every time point. Explored by Yutong He et al, and referred to as spatio-temporal super-resolution, it is of particular interest in the field of cinematography [1]. The situation examined in that study uses a combination of old and new satellite imagery to upscale low-resolution imagery, with excellent results. I see this as being useful in the video remastering process, where workers often have older, low-resolution footage which needs to be modernised.