Adaptive Regularized Blind Parametric Super-resolution Restoration and Enhancement
自适应正则有参超分辨率图像盲恢复

Image super-resolution restoration and enhancement (SR) based on reconstruction is a typically ill-posed and high-dimensional problem, which needs effective regularization to stable the solution. Lately a parametric and regularized blind SR( RPSR) was proposed by Nguyen et al, which has set up a frame work for the blind SR. Under the frame of RPSR, in this paper, an adaptive RPSR(ARPSR) based on image locale smoothing characteristics is put forward, and for the conveniences of computing, an approximate ARPSR is proposed also, by which at first the ARPSR problem is transformed into a weighted combination of two RPSR problems, then the optical blurring and regularization free parameters are estimated by the standard RPSR frame, and then by exploiting the structures of the reordered system matrices, a preconditioner is constructed for the preconditioned conjugate gradient method(PCG) by which the high-resolution image is solved at last. Computational analyses and experimental results with synthetic low-resolution sequences show the improvements of ARPSR to the RPSR frame.