Effective hyper-spectral image segmentation using multi-scale geometric analysis

The wide availability of hyper-spectral (HS) images has fostered the development of new algorithms for remote sensing applications ranging from agricultural and environmental to military use. Nevertheless, the analysis of such voluminous data requires advances analysis and computational methodologies as well as advanced hardware and computational methods. In this paper we introduce a new state of the art method for segmentation of hyper-spectral images. The proposed methodology is based on a multi-scale geometric transformation called the Beamlet Transform. The method is applicable for both mono-spectral and hyper-spectral images where each pixel has its corresponding spectral profile vector. The proposed segmentation method is especially effective when the underlying image consist of relatively large segment with smooth boundaries. In this case it performs exceptional well even in extremely low SNR. The method is unsupervised and assumes no prior knowledge of the image characteristics or features. Furthermore, it involves free parameters which allow fine tuning for a specific application, improving segmentation results. In order to validate the efficiency of our method we used the known Lark algorithm as a benchmark for segmentation of multi-spectral images and show that our new method out-performs the Lark algorithm.