Title :
Contextual unmixing of geospatial data based on Markov random fields and conditional random fields
Author :
Nishii, Ryuei ; Ozaki, Tomohiko
Author_Institution :
Fac. of Math., Kyushu Univ., Fukuoka, Japan
Abstract :
In supervised and unsupervised image classification, it is known that contextual classification methods based on Markov random fields (MRFs) improve the performance of non-contextual classifiers. In this paper, we consider the unsupervised unmixing problem based on MRFs. The exact solutions maximizing local conditional densities are derived, and they show excellent performance for unximing of data sets. Furthermore a new stochastic model based on conditional random fields is proposed for unmixing of hyperspectral data. The approximation formula of its normalizing factor is also derived.
Keywords :
Markov processes; image classification; random processes; Markov random field; conditional random field; contextual classification; contextual unmixing; geospatial data; hyperspectral data; local conditional densities; noncontextual classifier; normalizing factor; stochastic model; unsupervised image classification; unsupervised unmixing problem; Hyperspectral imaging; Image classification; Logistics; Machine learning; Markov random fields; Mathematics; Multispectral imaging; Pixel; Stochastic processes; Training data; CRF; Dirichlet distribution; MRF; logistic discrimination; unmixing;
Conference_Titel :
Hyperspectral Image and Signal Processing: Evolution in Remote Sensing, 2009. WHISPERS '09. First Workshop on
Conference_Location :
Grenoble
Print_ISBN :
978-1-4244-4686-5
Electronic_ISBN :
978-1-4244-4687-2
DOI :
10.1109/WHISPERS.2009.5289004
