File(s) under permanent embargo
Spectral unmixing based on nonnegative matrix factorization with local smoothness constraint
conference contributionposted on 2015-01-01, 00:00 authored by Z Yang, L Yang, Z Cai, Yong XiangYong Xiang
Spectral unmixing (SU) is an emerging problem in the remote sensing image processing. Since both the endmember signatures and their abundances have nonnegative values, it is a natural choice to employ the attractive nonnegative matrix factorization (NMF) methods to solve this problem. Motivated by that the abundances are sparse, the NMF with local smoothness constraint (NMF-LSC) is proposed in this paper. In the proposed method, the smoothness constraint is utilized to impose the sparseness, instead of the traditional L1-norm which is restricted by the underlying column-sum-to-one requirement of the to the abundance matrix. Simulations show the advantages of our algorithm over the compared methods.