Image and Video Restoration with TV/L2-Norm Constraint

There is a high demand for generating high-quality video and images, which are used for the wide range of applications, such as biometric authentication, medical imaging, and so on. In this paper, we present a video restoration method for generating a high-quality video from a deteriorated or blurred video. Recent researches independently investigate how to estimate the proper kernel from single blurred image and other researchers developed the unique algorithm, which includes time variation of data for restoring a blurred video. Therefore, in this paper we proposed a high-quality video or image restoration method, which combined with these two researches' methods for enhancing restoration performance. Our strategy can be divided into two steps. The first step is kernel estimation from each image frame of blurred video with using Total Variation (TV)/L2-norm as a regularization term. Then, the second step is to recover a high-quality video with algorithm, which considering time variation of adjacent frames. Experimental results show that the recovered high-resolution video and images with our proposed approach can achieve comparable performance than the conventional methods. In addition, our method can visualize how the blurry degree changes in the video.