Image Super-Resolution Using DWT Based Learning with Zooming Approach
Meera Doshi, Prakash P. Gajjar, Ashish M. Kothari · 2018
In this paper, we suggest a novel approach of learning based procedure for conserving fine details of low resolution image which are observed at different camera zoom-lens. We acquire a picture of complete scene at a resolution equivalent to the most zoomed observation using the images of a static scene recorded with three distinct zoom factors. This is called multi-frame super-resolution. The spatial resolution of low resolution image is enhanced by learning features from high resolution images of the training dataset. Training dataset consists of low resolution and corresponding high resolution images records. The resultant high-resolution image is modelled through the fact that crucial information of image such as edges, corners, and curves in the low resolution images are almost identical to their high resolution images. To retrieve super resolved image, we use four neighbourhood pixels comparison method and employ discrete wavelet transform to capture detailed information from training dataset and to learn high frequency coefficients of high resolution training images. The high resolution versions of low resolution observations at different camera zooms are combined. The suggested approach has been trialled on numerous real world natural images. The outcomes show that the suggested approach is better in both qualitative and quantitative manner over existing approaches.