A Cubic B-Splines Approximation Method Combined with DWT and IBP for Single Image Super-resolution

Abstract

The process of converting low-resolution images into high-resolution images by removing noise and estimating high-frequency information is known as image super-resolution. Aliased and decimated versions of the actual scenes are considered low-resolution images. The edges of high-resolution images produced by super-resolution from a single image are typically blurred. This paper proposes an approach to generate high-resolution image with sharp edges by combining a cubic B-Splines approximation, a discrete wavelet transform (DWT), and an iterative back-projection (IBP) edge-preserving weighted guided filter. A two-stage cubic B-Splines approximation, which includes pre-filtering and interpolation, is employed to up-sample the low-resolution image. The pre-filtering approach is used to transform pixel values to B-Splines coefficients. This approach minimizes blurring in the up-sampled image. The lost high-frequency information is then estimated using a one-level discrete wavelet transform based on the db1 wavelet. Finally, using a weighted guided filter, the resulting image is subjected to back-projection to obtain a high-resolution image. The proposed single-image super-resolution approach is applied on RGB colour images. The proposed method outperforms other selected approaches for comparison objectively in terms of PSNR and SSIM and also in visual quality.