Improving image quality using deep learning based super resolution
Abstract – Super resolution is one of the important computer vision tasks. A low-definition image can be changed to the high-definition image through super resolution. Likewise, it can be applied to a video. Especially, the popularization of smart devices and the inundation of video-based contents cause a gradual increase in the importance of super resolution tasks. However, the super resolution task is an ill-posed problem without only one correct answer. The reason is that given a low-definition image, there is no only one answer sheet corresponding to the high-definition version of the corresponding image with possible multiple answer sheets. Namely, super resolution is the technology greatly contributing to the society but has an ill-posed problem. Hence, super resolution is a very invaluable subject from the research perspective. The study performs the recurrence of the existing methodology for super resolution and presents the new deep learning model called the Boosted Super Resolution Generative Adversarial Nets (BSRGAN) by improving the methodology.