Super Resolution Image Reconstruction

Abstract

Super Resolution is a process to increase the resolution of an image using information from several different images. Each low resolution images has new information of the image and the main aim of super resolution is to combine these low resolution images to enhance the image resolution. In this thesis work, evaluation of different reconstruction based super resolution algorithms in order to enhance the low resolution images is done by experiments using simulated and real low resolution images. The observation model generates different simulated low resolution images obtained by rotation, translation, blurring and adding noise on high resolution image. For motion estimation different image registration algorithms (Vandewalle et al, Marcel et al and Keren et al) are applied to those simulated low resolution images. The parameters obtained from image registration are used to reconstruct the high resolution on different image reconstruction algorithm (Interpolation, Papoulis Gerchberg, Iterated Back Projection (IBP), Robust Super Resolution, Projection onto Convex Sets (POCS)). The obtained images are passed through Median filter and Weiner filter to remove blur and noise. The result is then compared and analyzed by means of objective image quality criteria mean square error(MSE) and peak signal to noise ratio(PSNR). The overall process is repeated for reconstruction of the real low resolution images. Finally, the obtained result is examined to show the performance of algorithms. Keywords: Super Resolution, Motion Estimation, Image Registration, Image Reconstruction