Endmember extraction by L_2,0 constrained sparse dictionary selection

Endmembers play an important role in many hyperspectral remote sensing applications, such as classification and Spectral Mixture Unmixing (SMU). In this paper, by considering endmembers as a small subset of pixels in a hyperspectral image, a sparse Linear Mixture Model (sLMM) is constructed to model the mixed pixels. As a result, an L_2,0 based sparse dictionary selection model is proposed for endmember extraction (EE) of hyperspectral images. Moreover, a Simultaneous Orthogonal Matching Pursuit (SOMP) based algorithm is adopted to extract endmembers efficiently. Experimental results on both synthetic and real hyperspectral data demonstrate our proposed EE algorithm outperforms several popular pure-pixel EE algorithms.