Efficient image magnification and applications to super-resolution reconstruction

Image magnification, or interpolation, produces a high resolution image from a low resolution, and perhaps noisy image. There have been proposed a variety of magnification works. However, they are either sensitive to the noise, or non-robust to the blocking artifacts, or of high computational complexity, which hence limits their utility. In this paper, we propose an alternative magnification approach utilizing a filtering-based implementation scheme and novel regularization through coupling bilateral filtering with the digital total variation model. The method is simple, fast, and robust to both noise and blocking artifacts. Another novelty in the paper is the application of the novel regularization to super-resolution reconstruction, which leads to a new super-resolution method. Experiment results demonstrate the effectiveness of our method.