Fast segmentation of hyperspectral images by combining textural and spectral information
Abstract
In this paper different approaches for dimensionality reduction of hyperspectral images are compared in order to apply a texture based segmentation algorithm which efficiently operates in low dimensions. Detection of Powdery Mildew infection is used as a real-world test case for the proposed approach. For this application, an adaptive Principal Component Analysis based projection preserves the most relevant spectral information within the combined process of spatial-spectral feature extraction.
