Fusion of Multispectral and Panchromatic Images by Integrating Standard PCA with Rotated Wavelet Transform

Abstract

Many pansharpening algorithms are based on the principle of extracting spatial details from panchromatic (PAN) images and injecting them into multispectral (MS) images. In this paper, we present two fusion approach based on same principle by integrating standard principle component analysis (PCA) with decimated and undecimated rotated wavelet transform. When decimated/subsampled rotated wavelet transform (SSRWT) is used for fusion of MS and PAN images, three visual artifacts get introduced in the fused image namely color distortion, shifting effect and shift distortion. To eliminate color distortion, SSRWT is integrated with standard PCA, i.e., PCA–SSRWT. Color distortion is significantly mitigated, but shifting effect and shift distortion persist in the fused image of PCA–SSRWT. After employing undecimated/nonsubsampled rotated wavelet transform (NSRWT), shifting effect and shift distortion get eliminated with minimum color distortion. However, fused image as a result of NSRWT is spectrally high but spatially low. In order to improve spatial quality and remove visual artifacts observed in SSRWT and PCA–SSRWT, NSRWT is integrated with standard PCA, i.e., PCA–NSRWT. Visual and quantitative analysis is carried out to validate the quality of fused image for all the algorithms. Visual interpretation suggests that fused image obtained using PCA–NSRWT is superior to fused images of SSRWT, PCA and NSRWT. The overall quantitative analysis manifests that the PCA–NSRWT is consistent with visual interpretation and performs better than state-of-the-art methods. PCA–NSRWT not only removes visual artifacts but also improves spectral and spatial quality of the fused image compared to individual PCA, SSRWT, NSRWT and PCA–SSRWT. Based on visual and quantitative analysis, it is observed that PCA works better with undecimated compared to decimated rotated wavelet transform for fusion.