Grouplet-based color image super-resolution

This paper addresses the problem of generating a super-resolution (SR) image from a single multi-valued low-resolution (LR) input image. The main application in our case lies in the exploitation of the cinema or TV archives for projections in higher resolutions (HD, 2K, 4K). We approach this problem from the perspective of image geometry-oriented interpolation. First, the geometry of the LR image is obtained by computing the grouplet transform. The grouplet orthogonal bases, that were introduced by Mallat in [1], are constructed with a multiscale association field that groups pixels to take advantage of geometrical image regularities. These bases are used to define a grouplet-based structure tensor in order to capture the geometry and directional features of the LR color image. Then, the SR image is synthesized by an adaptive directional interpolation using the extracted geometric information to preserve sharpness of edges and textures. This is accomplished by the minimization of a functional which is defined on the extracted geometric parameters of the LR image and oriented by the geometric flow defined by the grouplet transform. The proposed super-resolution algorithm outperforms the state-of-the-art methods in terms of the visual quality of the interpolated image.