Performance analysis of wavelet filter bank for an image super resolution algorithm

Abstract

Image super-resolution is a technique in which a high-resolution image is generated using a single or multiple low-resolution images. In this paper, an image super-resolution algorithm is proposed in which Discrete wavelet transform (DWT) is used to generate different frequency sub-bands of the image and Stationary wavelet transform (SWT) overcomes the issue of lack of translation invariance of DWT so it is used here with DWT. To preserve more edge information Canny Edge extraction operator has been applied to the input image and subbands are interpolated using Lanczos interpolation. The high-frequency sub bands and the input image are passed through Non-Local Mean (NLM) filter to reduce the artifacts generated by DWT. Different orthogonal and bi-orthogonal filters have been applied to this algorithm and different quality parameters such as PSNR, MSE, RMSE, SSIM and Correlation coefficient are calculated. It is found that db2 wavelet is showing better results.