Classification of hyperspectral images by tensor modeling and additive morphological decomposition

• Published in: Pattern recognition Vol. 46; no. 2; pp. 566 - 577
• Main Authors: Velasco-Forero, Santiago, Angulo, Jesus
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.02.2013; Elsevier
• Subjects: Additives; Algorithms; Applied sciences; Artificial intelligence; Classification; Computer Science; Computer science; control theory; systems; Decomposition; Engineering Sciences; Exact sciences and technology; Hyperspectral images; Information, signal and communications theory; Mathematical analysis; Mathematical morphology; Pattern recognition; Pattern recognition. Digital image processing. Computational geometry; Pixelwise classification; Signal and Image processing; Signal processing; Spectra; Telecommunications and information theory; Tensor modeling; Tensors; Hyperspectral images; Pixelwise classification; Mathematical morphology; Tensor modeling; Performance evaluation; Pixelwise; Pattern recognition; Algorithm; Modeling; Statistical method; Hyperspectral imaging sensor; Dimension reduction; Multiscale method; Tensor calculus; Principal component analysis; Hyperspectral; Morphology
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2012.08.011

Pixel-wise classification in high-dimensional multivariate images is investigated. The proposed method deals with the joint use of spectral and spatial information provided in hyperspectral images. Additive morphological decomposition (AMD) based on morphological operators is proposed. AMD defines a scale-space decomposition for multivariate images without any loss of information. AMD is modeled as a tensor structure and tensor principal components analysis is compared as dimensional reduction algorithm versus classic approach. Experimental comparison shows that the proposed algorithm can provide better performance for the pixel classification of hyperspectral image than many other well-known techniques. ► Additive morphological decomposition without any loss of information is proposed. ► Decomposition is modeled as a tensor structure. ► Tensor PCA is compared versus PCA. ► Proposed workflow performs better than other techniques in HSI classification.