A novel graph-based approach for segmenting hyperspectral images which guarantees similarity of pixels (spectra) inside each segment.
A study was undertaken to find a divergence measure that can with some level of accuracy identify whether a pair of samples are similar or dissimilar. The algorithm then creates a segmentation such that all pixels inside the segment are similar/uniform. Following this we show that the new technique improves on the segmentation performance of existing algorithms.
This work is done by Arun Saranathan with Mario Parente and has been accepted to IEEE Transactions on Geoscience and Remote Sensing.
A. M. Saranathan and M. Parente, ”Uniformity-based superpixel segmentation of hyperspectral images,” IEEE Trans. Geosci. Remote Sens., vol. 54, no. 3, pp. 1419-1430, Feb. 2016 [link].