Pansharpening with support vector transform and semi-nonnegative matrix factorization

This paper attempts to reduce the spectral distortion and enhance the spatial information of fused images. For this purpose, the author presented a novel pansharpening method based on support vector transform (SVT) and semi-nonnegative matrix factorization (semi-NMF). The proposed method involves three steps. In step one, SVT was performed on panchromatic and multispectral images. In step two, the low-frequency components were processed by semi-NMF-based fusion rule, while the high-frequency components were treated by the regional energy-weighted fusion rule. In step three, the fused images were reconstructed by the fused high-frequency and low-frequency components. After that, the proposed method was compared with other related methods through experiments on several datasets collected from QuickBird and GeoEye-1. The comparison shows that the proposed method outperforms the compared approaches. The research findings shed new light on the preservation of spatial and spectral information in image fusion.