Regional vs. Global Superpixel-Based Unmixing of Hyperspectral Imagery

Abstract

In this paper, a superpixel-based hyperspectral image unmixing is implemented. First, superpixel segmentation is applied to the image. A low dimensional representation of the image is created by representing each superpixel by its mean spectra. Second, the low dimensional image is segmented into homogeneous tiles using quadtree approach. Endmember extraction is applied to each tile to extract local endmembers. Abundances estimation is performed on the full image using the extracted endmembers. The approach is compared with three different unmixing scenarios that incorporate low dimensional and spatial information. Experimental results are presented using the ROSIS Pavia University hyperspectral image.