Image classification using spectral and spatial information based on MRF models

Abstract

A new criterion for classifying multispectral remote sensing images or textured images by using spectral and spatial information is proposed. The images are modeled with a hierarchical Markov Random Field (MRF) model that consists of the observed intensity process and the hidden class label process. The class labels are estimated according to the maximum a posteriori (MAP) criterion, but some reasonable approximations are used to reduce the computational load. A stepwise classification algorithm is derived and is confirmed by simulation and experimental results.

An MRF-based kernel method for nonlinear feature extraction

“…The Markov random field (MRF) [32,29] models spatial context information by confining the mutual influences among sites within the neighborhood. Let H s denote the set of sites neighboring s. A random field F on a lattice T is said to be an MRF if and only if for any site s 2 T, the local Markovianity…”

“…The proposed kernel function is a composite kernel that integrates two additional resources of information -class membership and spatial context -while preserving the concept of the kernel trick. Spatial context information is explored by means of Markov random field models (MRFs) [32,33]. Two pixels are compared based on the characteristics of the spatial neighborhood.…”

Improving Potts MRF Model Parameter Estimation in Image Analysis

“…The standard Potts model assumes that the spatial parameter is globally constant throughout the field and it can be equivalently defined by a Gibbs joint distribution or by a set of local conditional density functions (LCDF's) [2]. For a general neighborhood system N, the LCDF of a Potts pairwise interaction model is [23]:…”

Improving forest type classification using the vegetation local difference index

“…Information on the methods and factors to consider when conducting image classifications can be found in detail in the review by Lu and Weng (Lu and Weng 2007). Several earlier studies on land-use/land-cover (LULC) classification have demonstrated that integrating spatial context and spectral information may provide additional information for classification and may increase the discrimination of LULC categories that improve the classification accuracy (Yamazaki and Gingras 1995;Kontoes and Rokos 1996;Li and Narayanan 2004;Jimenez et al 2005;Bandyopadhyay 2005;Ota, Mizoue, and Yoshida 2011). Moreover, the aforementioned applications explore the spatial information of highresolution images to improve the classification accuracy.…”