A new adaptive Bayesian super-resolution reconstruction algorithm for digital image sequence with different noise levels

A novel super-resolution reconstruction algorithm based on multiple low resolution images is proposed. Under this algorithm, various weights and regularization parameters are assigned to each low resolution image, and in the resolution process, weights and regularization parameters are updated in each iteration. The non-liner weights function is able to keep the weights stable in each image, and take full advantages of the effective information in the low resolution images. The optimal solutions are obtained by the relaxation iteration method, and the image is reconstructed. This algorithm is suitable for the inner independence or static state models. The experimental results show that the frames reconstructed by present algorithm are better than those from the cubic B-spline interpolation and the traditional Bayesian MAP estimation.