AN APPROACH FOR IMAGE SUPER RESOLUTION BASED ON KERNEL LEARNING

Abstract

Image super resolution is the most attractive research area to obtain high-resolution image using its low-resolution image
observation. Image super resolution reconstruction is an ill-posed problem, So some effective prior information is needed for
regularizing the super resolution problem and avoid infinite solution and enhance edges and suppress artifacts and generate high
quality super resolution solution while to avoid unexpected artifacts and patch redundancy in learning-based method. Reconstruction
and learning-based super-resolution methods are for restoring a high-resolution image from low-resolution images. The proposed
system introduce local prior suppress artifact by using steering kernel regression to estimate local gradient between neighboring values
and local orientation information. The proposed method tries to improve the performance so we can get the more optimized resolution
and avoid over smoothing of HR image. Experimental results show that proposed method produce large PSNR and SSIM value that of
state-of –the art approaches.