Random Set Framework for Context-Based Classification With Hyperspectral Imagery

Foreword to the Special Issue on Hyperspectral Image and Signal Processing

“…In a recent trend, contextual information is also taken into consideration to improve the classification performance obtained with only spectral information (pixelwise classification). These methods, which are referred to as spectral-spatial classifiers, typically focus on local spatial information and are particularly successful for data with large homogeneous regions or where spectral signatures of multiple classes overlap [68]- [72]. In this special issue, eight papers directly address the problem of classification and propose new algorithms.…”

Classification of hyperspectral images by tensor modeling and additive morphological decomposition

“…where P i = U i U i T , and U i is found by using expression (11). This representation allows to include noise filtering in the sense of SVD filtering [41] if only the largest eigenvectors are considered per mode.…”

“…Recent works in HSI have seen a surge of research toward developing approaches that exploit various features specific to the spatial/spectral classification. The approaches due to ( [7,8,9,10,11,12,13,14,15,16,17]) show some degree of success. Pixelwise classification incorporating spatial information in HSI can be roughly divided according to their mathematical formulation as follows.…”

“…• Context-based classification [11,8] attempts to identify relevant models to a test sample through context estimation in the feature space, by using random set framework [11] or hierarchical segmentation [8].…”

Unsupervised and supervised classification of hyperspectral imaging data using projection pursuit and Markov random field segmentation

“…The methodology of considering spatial information along with spectral information to improve the classification accuracy of a hyperspectral image has been a recent trend (Jackson and Landgrebe 2002, Fauvel et al 2008, Bolton and Gader 2009, Chen et al 2009b, Velasco-Forero and Manian 2009). Our approach proceeds on similar lines by combining spectral and spatial information to develop a classification technique that yields high classification accuracy vis-á-vis computational efficiency.…”