Hyperspectral unmixing is a very important task for remotely sensed hyperspectral data exploitation. It amounts at estimating the abundance of pure spectral signatures (called endmembers) in each mixed pixel of the original hyperspectral image, where mixed pixels arise due to insufficient spatial resolution and other phenomena. A challenging problem in spectral unmixing is how to automatically derive endmembers from hyperspectral images, particularly due to the presence of mixed pixels which

more »

... erally prevents the localization of pure spectral signatures in transition areas between different land-cover classes. A possible strategy to address this problem is to guide the endmember extraction process to spatially homogeneous areas. For this purpose, several preprocessing methods (intended to be applied prior to the endmember extraction stage) have been developed in the literature. However, most of these methods only include spatial information during the preprocessing and disregard spectral information until the subsequent endmember extraction stage. In this paper, we develop a new joint spatial and spectral preprocessing method which can be combined with any endmember extraction algorithm from hyperspectral images. The proposed method is intended to retain spectrally pure pixels which belong to spatially homogeneous areas. Our assumption is that spectrally pure signatures are more likely to be found in spatially homogeneous areas rather than in transition areas between different land-cover classes, which are expected to be dominated by mixed pixels. Our experimental results, conducted with a variety of hyperspectral images, reveal the robustness of the proposed method when compared to other similar preprocessing strategies.