Image enhancement algorithms for low resolution images using evolved wavelet filters

Abstract

Satellite images are more frequently used nowadays and the resolution with which it is available to the user is limited. Hence it necessitates the need for robust resolution enhancement algorithm which works well for satellite images. The interpolation algorithms like nearest neighborhood, bilinear and bicubic algorithm works considerably well for natural images. But, there is a need for high level image enhancement algorithms which can work well for satellite images with more edges. In this research work, an image resolution enhancement algorithm based on evolved wavelet-filter coefficients is proposed for both satellite and medical images. The main focus of this research work lies in the optimization of wavelet filters based on parameters related to edges in the image. Initially all the input images are classified into various bins based on the edges present in the image and each bin of images will be enhanced by applying unique wavelet filters. These libraries of wavelet filters are evolved using bioinspired algorithm like genetic algorithm considering individual image bins. The individual image groups of satellite images are created based on the Spatial Frequency Mean (SFM) and wavelet filters are evolved for each group for both near edge and far edge image namely local and global DWT (CDF 9/7 type) filters. Two enhancement algorithms namely SDWT (Standard DWT) and EDWT (Evolved DWT) based on existing DWT-based methods are proposed to improve the resolution of the low-resolution satellite and medical images. The proposed algorithm enhances the resolution of the satellite image and it shows a significant improvement in the PSNR value of 0.5 dB compared with the existing techniques. The proposed algorithm is validated using 100 test satellite images. Also, the same hybrid algorithms are validated for medical images where it shows a significant improvement in PSNR and SSIM. The proposed algorithms for medical images show a maximum PSNR improvement of 4dB. Also, the proposed hybrid algorithm 2 shows a significant and maximum improvement in SSIM than algorithm 1 newline