THRESHOLD BASED SEGMENTATION METHOD FOR HYPERSPECTRAL IMAGES

Superpixel algorithms attempt to group contiguous image pixels which are in homogeneous regions into segments(superpixels). For hyperspectral images the pixels contained in a segment must exhibit spectral uniformity, namely they must have similar spectral shapes. Superpixel representation have been used to improve the performance of unmixing algorithms on hyperspectral images. Superpixel algorithms applied to hyperspectral images provide no guarantees on the uniformity inside each segment. The absence of such guarantees requires an over segmentation of the image to maintain spectral variablity. We introduce a graph based agglomerative approach for superpixel segmentation that will provide some basic guarantees on the uniformity inside a segment, which can be used to push the segmentation. We will show that these segmentations with uniformity guarantees can be used to generate mineralogical summaries for images in the Compact Reconnaissance Imaging Spectrometer(CRISM) data-set and the Airborne Visible/Infrared Imaging Spectrometer(AVIRIS) data set.