Sparse Regularization based Fusion Technique for Hyperspectral and Multispectral Images using Non-linear Mixing Model

摘要

In this paper, an image fusion technique for fusing hyper spectral and multispectral images based on sparse regularization and subspace modeling is proposed. Here, the problem of fusion is modeled as a linear inverse problem and is solved in a lower dimensional subspace. Non Linear Mixing Model (NLMM) of hyper spectral image is used for the subspace identification and it gives better results than Linear Mixing Model (LMM). A sparse regularization term is generated through adaptive dictionary learning and the fusion task is solved by using alternating optimization technique. Subspace modeling reduces computational complexity considerably. Experimental results show that this method offers significant improvement in fusion performance when compared to that of existing methods.