Automated selection of results in hierarchical segmentations of remotely sensed hyperspectral images

Abstract

Abstract : The hierarchical image segmentation (HSEG) algorithm is a hybrid of hierarchical step-wise optimization and constrained spectral clustering. Unlike most other segmentation approaches, HSEG produces a hierarchical set of image segmentations. A single segmentation level can be selected out of the segmentation hierarchy by examining how the features or individual regions change throughout the different levels of detail. Subsequently, the selection of a single segmentation result for each region can effectively transform the segmentation hierarchy into a region-adaptive segmentation approach. The above task has previously been accomplished using supervised and time-consuming procedures. This paper presents a first step towards the automation of this process, where spatial, spectral and joint spectral/spatial features are used to investigate how regions change from one hierarchical level to the next for region identification in remotely sensed hyperspectral data sets. Comparative results are presented using Airborne Visible-Infrared Imaging Spectrometer (AVIRIS) data collected over the Salinas Valley in California.