A new maximum distance method based on barycentric coordinate for endmember extraction

摘要

In this paper, we present a new maximum distance method (MDM) by applying the simplex barycentric coordinate for endmember extraction, named BC-MDM. Under linear mixing model and pure-pixel assumption, the MDM, which is proposed by Geng in 2005, is based on the fact that a pixel is an endmember if it has the maximum distance to the sub-simplex with vertices being endmembers already extracted. Calculating the distance of a pixel to the sub-simplex is quite time-consuming. To overcome this problem, we introduce the idea of barycentric coordinate to MDM, which transforms the problem of distance calculation into spectral unmixing with abundance sum-to-one constraint. As a result, BC-MDM has greatly reduced the computational complexity of original MDM. The experimental results with real hyperspectral image demonstrate that compared to the simplex growing algorithm (SGA), the proposed method can provide the same performance with a computational complexity between one and two orders magnitude lower, and compared to vertex component analysis (VCA), our method can provide consistent result while requires less computing time.