Interpolation based on similarity in subband generation for image super resolution

A super resolution system aims at producing artifact and noise free higher resolution images from lower resolution ones, with interpolation being its pivotal part. While generating a higher resolution image, the interpolation phases in most super resolution techniques introduce different artifacts such as blurring and ringing. In this paper, we propose a technique where the interpolation is performed through estimation of high frequency components by optimally exploiting the across-scale similarity in the process of wavelet decomposition of the input image. The decompositions considered are obtained by suitably fusing outputs of discrete (DWT) and stationary wavelet transforms (SWT). The result shows that our approach is considerably faster than the state-of-the art. Qualitative analyses reveal that it not only avoids artifacts but also makes edges and finer details prominent. In terms of PSNR and SSIM, the proposed approach is found to provide performance near to the best in significantly less computational time.