Adjacency selection in Markov Random Fields for high spatial resolution hyperspectral data |
| |
Authors: | Francesco Lagona |
| |
Institution: | (1) Department of Social Sciences, University “Roma Tre”, via Corrado Segre 4, 00146 Rome, Italy (e-mail: lagona@uniroma3.it), IT |
| |
Abstract: | Markov Random Fields, implemented for the analysis of remote sensing images, capture the natural spatial dependence between
band wavelengths taken at each pixel, through a suitable adjacency relationship between pixels, to be defined a priori. In most cases several adjacency definitions seem viable and a model selection problem arises. A BIC-penalized Pseudo-Likelihood
criterion is suggested which combines good distributional properties and computational feasibility for analysis of high spatial
resolution hyperspectral images. Its performance is compared with that of the BIC-penalized Likelihood criterion for detecting
spatial structures in a high spatial resolution hyperspectral image for the Lamar area in Yellowstone National Park.
Received: 9 March 2001 / Accepted: 2 August 2001 |
| |
Keywords: | : Adjacency selection Bayesian penalization rate hyperspectral data pseudo-likelihood Markov random fields |
本文献已被 SpringerLink 等数据库收录! |
|