Endmember extraction using a novel Cluster-based Spatial Border Removal Preprocessor

Most algorithms applied for extracting endmembers utilize spectral content of pixels with inattentive to spatial arrangement between them. Recently Spatial Spectral Preprocessor (SSPP) has been proposed for solving this problem. In this paper, we propose a novel Cluster-based Spatial Border Removal Preprocessor (CSBRP) by removing mixed pixels located at spatial borders of cluster map and calculate spectral purity weight of residual pixels in order to look for spectrally pure pixels thorough them so that the best pure pixels are adopted for the next EE stage. The performance of our method is appraised on a synthetic image derived by Rterrain HYDICE dataset and USGS library from the viewpoints of RMSE reconstruction, average minimum SAD and total processing time. Results relatively outperform the state-of-the-art techniques especially in low signal to noise ratio.