Image Superresolution with Multi-Kernel Estimation Embedded in Iterative Backward Projecting

The dominating methods in image super-resolution are mostly based on Bayes estimation which is usually completed by Iterative Backward Projecting. Under some special imaging condition, multi-kernel estimation for image interpolation based on training works well. Considering image quality degradation due to sampling process, in this paper, we incorporate the multi-kernel interpolation estimation into the computational frame of Iterative Backward Projecting (IBP) process. Our results demonstrate that the new method is superior to either multi-kernel estimation or IBP used individually.