The research on effectiveness of spectral similarity measures for hyperspectral image

Hyperspectral images have considerable spectral information, and lots of spectral similarity measures have been developed for hyperspectral image analysis. However, little research has been done on the effectiveness of the spectral similarity measures. This paper introduced three spectral discriminatory measures, the spectral discriminatory probability, the discriminatory power and the spectral discriminatory entropy as the objective statistical criteria. The performances of the four spectral similarity measures, i.e. the Euclidian distance (ED), the spectral angle measure (SAM), the spectral correlation measure (SCM), the spectral information divergence (SID), were evaluated on the AVIRIS image data set. The experiment results showed that the SID and SCM can better discriminate two spectra and have better chance to identify a target spectrum via the known spectral library. The experiments also reflect that SID and SCM have little influence by the noise.