ABSTRACT
EMPs Extended morphological pro¦les EMPs Extended morphological pro¦les LDA Linear discriminant analysis LogDA Logarithmic discriminant analysis MLR Multinomial logistic regression MLRsubMRF Subspace-based multinomial logistic regression
followed by Markov random ¦elds MPs Morphological pro¦les MRFs Markov random ¦elds PCA Principal component analysis QDA Quadratic discriminant analysis RHSEG Recursive hierarchical segmentation ROSIS Re©ective optics spectrographic imaging system SVMs Support vector machines TSVMs Transductive support vector machines
Hyperspectral imaging is concerned with the measurement, analysis, and interpretation of spectra acquired from a given scene (or speci¦c object) at a short, medium, or long distance, typically, by an airborne or satellite sensor [1]. Ÿe special characteristics of hyperspectral data sets pose di§erent processing problems, which must be necessarily tackled under speci¦c mathematical formalisms [2], such as classi¦cation and segmentation [3] or spectral mixture analysis [4]. Several machine learning and image-processing techniques have been applied to extract relevant information from hyperspectral data during the last decade [5,6]. Taxonomies of hyperspectral
image-processing algorithms have been presented in the literature [3,7,8]. It should be noted, however, that most recently developed hyperspectral image-processing techniques focus on analyzing the spectral and spatial informations contained in the hyperspectral data in simultaneous fashion [9]. In other words, the importance of analyzing spatial and spectral information simultaneously has been identi¦ed as a desired goal by many scientists devoted to hyperspectral image analysis. Ÿis type of processing has been approached in the past from various points of view. For instance, several possibilities are discussed by Landgrebe [10] for the re¦nement of results obtained by spectral-based techniques through a second step based on spatial context. Such contextual classi¦cation [11] accounts for the tendency of certain ground cover classes to occur more frequently in some contexts than in others. In certain applications, the integration of high spatial and spectral information is mandatory to achieve su¶ciently accurate mapping and/or detection results. For instance, urban area mapping requires su¶cient spatial resolution to distinguish small spectral classes, such as trees in a park or cars on a street [12] (Figure 12.1).
