Super resolution image reconstruction using averaged image and regularized deconvolution

Super resolution (SR) image reconstruction is used to obtain a high resolution (HR) image from multiple low resolution (LR) images. In this paper, we propose a new method to solve the SR reconstruction problem. The LR images are modeled as downsampled images of the original scene shifted by sub-pixel distances. In addition, we model the downsampling process as averaging light intensity on the corresponding pixel area. Based on this downsampling model, the average of multiple LR images with appropriate registration can be thought of as a blurred version of the HR image. After the point spread function (PSF) for this blur is identified, the HR image is obtained by regularized deconvolution method. The regularization factor can be determined by line search of a cost function. Two distinct cases are considered: (1) only translational motion among LR images is assumed, (2) both translational and rotational motions among LR images are considered. For the second case, the rigid body group representation is used.