Super-Resolution Method by Registration of Multi-scale Components Considering Color Sampling Pattern and Frequency Spectrum Power of UHDTV Camera
Yasutaka Matsuo, Shinichi Sakaida · 2016
We propose a video super-resolution method from 4K to 8K pixels to broadcast 4K video for future 8K broadcasting. 4K video has high self-similarity because it contains many similar objects in a single frame. Therefore, a super-resolution method by registration of multi-scale components in a single frame is proposed. Furthermore, the typical 4K camera has Bayer or other color sampling patterns in a one-CMOS sensor. For this reason, the multi-scale components used for the alignment procedure of this registration are selected by considering the color sampling pattern of the 4K camera. In addition, an assignment procedure using the alignment result of this registration is performed by spectrum power control that is used to estimate the power level of a super-resolved high-frequency band to prevent excessive emphasis. Experiments show that the proposed method provides an objectively better PSNR measurement and a subjectively better appearance in comparison with the conventional super-resolution methods.