SAR and visible image fusion based on local non-negative matrix factorization

Abstract

Although some of the traditional methods of image fusion such as wavelet transform fusion and Laplacian pyramid fusion have good effect on most visible images, it is not suitable for SAR image fusion. Because the speckle noise in SAR images is multiplicative and coherent. In this paper, we propose a method called local non-negative matrix factorization (LNMF) for SAR image fusion. LNMF uses multiplicative iteration to approximate the standard image and reduces speckle noise. For getting more localized, parts-based representation of images, LNMF improves the objective function of the standard NMF to enhance localization constraint. The result of experiments approved that LNMF method is efficient and effective for image fusion of SAR and visible images compared to other traditional methods.