Robust Nonnegative Local Coordinate Factorization for Hyperspectral Unmixing

Recently, nonnegative matrix factorization (NMF) has become increasingly popular for hyperspectral unmixing (HU). Due to the non-convex nature of the NMF theory, which is sensitive to the initial value and various noise. To obtain more accurate and robust unmixing model, in this paper, we propose a novel method called robust nonnegative local coordinate factorization (RNLCF). RNLCF adds a local coordinate constraint into the composite loss function which combing classic and Correntropy Induced Metric NMF objective function. To solve RNLCF, we developed a multiplicative update rules. Experimental results on synthetic and real-world data verify the effectiveness of RNLCF comparing with the representative methods.