Edge-Guided Single Depth Image Super Resolution

Abstract

In this paper Total variation is utilized as a prominent and effectual image prior model in the regularization based image processing fields. Nonetheless, as the total variation model supports a piecewise steady solution, this process comes under high intensity noise in the level areas of the picture is often poor, and a few pseudo edges are formed. In this work we develop a spatially adaptive total variation model. At first, the spatial information is extracted supported each and every pixel, and at that point 2 filtering process are added to restrain the impact of pseudo edges. In addition of this, the spatial info weight is built and classified with k-means clustering, and also the regularization strength in every region is controlled by center value of the cluster. The exploratory results, on both simulated and genuine datasets, demonstrate that the proposed methodology can adequately diminish the pseudo edges of the total variation regularization in the flat areas, and keep up the partial smoothness of the HR images. If we compare the traditional pixel based spatial information adaptive methodology, the proposed region based spatial information adaptive variation model can effectively reduce the effect of noise on the spatial data extraction and maintain strength with changes in the noise intensity in the SR process.