LAUSR

Abstract The prime intention of super-resolution (SR) technique is to restore the high-resolution images from one or more low-resolution (LR) images. These images are captured from the same scene with different acquisition systems with different resolution. Because these acquisition systems, images are suffered for an ill-posed problem with low visualization and picture information. Therefore, in this paper, the zoom-based super-resolution approach is proposed for super-resolution of low resolute images which are acquired from different camera zoom-lens. In this approach, three LR images of the same static scene which are acquired using three distinct zoom factors are used. Learning-based SR technique is used to enhance the spatial resolution of these LR images. The training dataset comprises three sets of captured images which are LR images, an enhanced version of LR images-HR1 and enhanced version of HR1 images-HR2. High-frequency details of the super-resolute image are learned in form of the discrete cosine transform (DCT) coefficients of HR training images. Finally, the super-resolved versions of LR observations, captured at different zoom-factors, are combined. The experimental results show that this proposed approach can be applied to various types of natural images in grayscale as well as color. The experimental results also show that this proposed approach performs better than existing approaches.