Joint spatial variables nonnegative matrix factorization using constrained gradient method to pansharpen multispectral images

Pansharpening is an important process in remote sensing that aims at combining observable panchromatic and multispectral images, to produce an unobservable pansharpened multispectral data. This fusion technique takes the advantage of the complementary spatial/spectral resolution characteristics of the considered images. In this paper, a new constrained gradient method for joint spatial-variables nonnegative matrix factorization is proposed to pansharpen multispectral images. In this method, a spatial degradation model between the two observable images is exploited in a joint spatial-variables criterion based on linear spectral unmixing concept. To optimize this criterion, new iterative gradient based update rules are designed. The proposed approach is applied to synthetic and real data. Its efficiency in both spatial and spectral domains is evaluated by commonly used performance criteria. Experimental results show that the proposed approach yields satisfactory spectral and spatial fidelities of the pansharpened images. Also, it outperforms tested pansharpening literature methods.