Graph Regularized L1/2-Sparsity Constrained Non-Negative Matrix Factorization for Hyperspectral and Multispectral Image Fusion

摘要

High spectral and high spatial resolution is paramount for high performance identification and classification of hyperspectal (HS) images. There are many different approaches in order to improve HS spatial resolution. Data fusion is the process of combining HS and multispectral (MS) images in order to obtain high spectral and high spatial resolution HS images. In this study, based on the coupled nonnegative matrix factorization (CNMF) framework for data fusion, L1/2-sparsity constrained graph regularized nonnegative matrix factorization (GLNMF) approach is investigated for HS and MS data fusion. Experimental results show that the GLNMF based fusion approach outperforms state-of-the-art CNMF based data fusion. Experimental results are illustrated on datasets synthesized according to Wald’ s protocol from AVIRIS Indian Pines and HYDICE Washington D.C. datasets.