Super resolution via generalized statistical smoothing

We constructed a technique to reconstruct a high-resolution original image using multiple low-resolution corrupted images for the framework of the reconstruction-based super-resolution utilizing image registration via the correlation method, resolution transformation via the bi-cubic method and noise reduction via the generalized statistical smoothing (GSS). Using numerical simulation for 256-grayscale standard images, we clarified that the present method achieves optimal performance for the reconstruction-based super-resolution, if we appropriately set the parameter for generalized parameter scheduling and threshold for edge enhancement in the GSS for noise reduction. Also, we found that the present method reconstructs an original image more accurately than that of the method using the correlation method, the bi-cubic method and the conventional filter, such as the average and Gaussian filter.