Region-Wise Super-Resolution Algorithm Based On the Viewpoint Distribution

Abstract

Recently, super-resolution techniques have been energetically studied for the purpose of reusing the low resolution image contents. Although a lot of approaches to achieve the appropriate super-resolution have been proposed such as non-linear filtering, total variation regularization, deep learning etc., the characteristic of the viewpoint distribution of the observer has not been effectively utilized. Because applying super-resolution to unimportant regions in an image may hinder the observer’s attention to seeing the display, it leads to a low subjective evaluation. This paper proposes the region-wise super-resolution algorithm based on the view-point distribution of observer. However, we cannot obtain the viewpoint distribution map for an image without the pre-experiment using the device such as eye mark recorder, therefore, the saliency map is utilized in this paper. Numerical examples show that the proposed algorithm using saliency map achieves a higher subjective evaluation than the previous study based on the non-linear filtering based super-resolution. Furthermore, in numerical examples, the proposed algorithm using the saliency map is shown to give the similar results of the algorithm using the viewpoint distribution map obtained by the pre-experiment using eye mark recorder.