Improving the Classification in Shadowed Areas using Nonlinear Spectral Unmixing

This paper presents a shadow restoration method based on the nonlinear mixture model. A shadowed spectrum is modeled by using a pure sunlit spectrum for the same material following physical assumptions. Regarding pure sunlit and shadowed spectra as endmembers, an unmixing process is then conducted pixel-wise using a nonlinear mixture model. Shadow pixels are restored by simulating their exposure to sunlight through a combination of selected sunlit endmembers spectra, weighted by abundance values. Experiments conducted on a real airborne hyperspectral image are eval- uated through spectra comparison and classification. In addition, a soft shadow map is generated, which quantifies the shadow intensity at the edges between sunlit and shadow areas