《Level Set Segmentation of Hyperspectral Images Using》.pdf

Level Set Segmentation of Hyperspectral Images Using Joint Spectral Edge and Signature Information Radford R. Juang+ Philippe Burlina+* Amit Banerjee+ Johns Hopkins University Applied Physics Lab+ Dept. of Computer Science* Laurel, MD 20723 radford.juang@jhuapl.edu Abstract - This paper describes a new method for Bayesian or data clustering approaches. Given an image, segmenting hyperspectral imagery (HSI) using dynamic these methods extract d-dimensional feature vectors from curves. We are concerned about challenging HSI target the data and map them into a d-dimensional feature space. segmentation/detection use cases where the scene Feature vectors from pixels of similar objects or materials includes confusers exhibiting a spectral return similar will form dense local clusters in the feature space. to the desired signature and in close proximity of the Segmentation is then achieved by detecting and object of interest. Our method is based on a level sets segregating these clusters. approach. It fuses all available spectral bands and Bayesian methods generally proceed by formulating incorporates spectral as well as spatial information to statistical model assumptions for the region generation obtain a finer target segmentation. The proposed and image formation processes. Maximum likelihood method applies level set segmentation to HSI by defining