Super-resolution video reconstruction based on both local and global information

Abstract

Although super-resolution (SR) methods have been successfully used to improve the resolution of video content, these methods estimate high resolution (HR) frames without explicitly use local information. Instead, they minimize the sum of difference between acquired low resolution (LR) images and observation model. On the contrary, adaptive kernel regression estimates each pixel of HR frames independently. It does not consider global optimum while estimating HR frames. In this paper, we proposed an idea of employing adaptive kernel regression on SR methods to improve the quality of super-resolved video frames. It is shown that the proposed idea can provide results with better visual quality and Peak Signal-to-Noise Ratio (PSNR).