Hyperspectral Gradient Analysis: A First Study

We present a physics-based gradient adapted for metrological analysis of surfaces using hyperspectral imaging. The multivariate image gradient is mathematically split into spectral shape gradient (SSG) and spectral intensity gradient (SIG). The obtained decomposition provides a direct access to material changes as well as a more robust spatial gradient direction assessment thanks to the SSG. The performances are demonstrated via watershed-based segmentation applied on well-known remote sensing images. Results show significant improvement over the state-of-the-art.