Face super-resolution via semi-kernel partial least squares and dictionaries coding

In this paper, a patch-based super-resolution (SR) method is proposed to hallucinate facial images. Two steps are involved in this method. In the first step, we combine semi-kernel partial least squares (semi-KPLS) algorithm with collaborative representation (CR) to infer an initial high-resolution (HR) face. In the second step, sparse representation in gradient domain is employed to compensate the global face with detailed inartificial facial features. Furthermore, the gradient images, obtained from sparse representation, are integrated with the gradient of initial HR face. Based on the integrated gradient images, a generalized Poisson solver is used to reconstruct the final high resolution image. The experiments conducted on FERET database demonstrate the proposed algorithm can generate better results, in comparison with some state-of-the-art approaches.