An Unsupervised Segmentation Method Using Markov Random Field on Region Adjacency Graph for SAR Images
- 1 October 2006
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-4
Abstract
A fast approach to obtain segmentation of SAR images has been suggested here based on the local statistical characteristics using Markov random field (MRF) model on region adjacency graph (RAG). First, an initially over-segmented image derived from the watershed segmentation algorithm as well as the original SAR image is taken as the inputs of the proposed method. Secondly, a MRF is defined on RAG of the initial over segmented regions, with a novel multilevel logistic (MLL) model for the region class labels and Gamma distribution for the marginal distribution of each class in the SAR images. The criterion used for getting the optimal segmentation is the maximization of the posterior marginal (MPM), which minimizing the expected value of the number of the misclassified regions in the over-segmented image. In the implementation, the expectation maximization (EM) algorithm is used to estimate the parameters of Gamma distribution, and the parameters of the MLL model is derived from the RAG. Experimental results on real SAR images show that the proposed method can reduce the computational complexity greatly and provide precise segmentation resultsKeywords
This publication has 13 references indexed in Scilit:
- An Unsupervised Segmentation Method Based on MPM for SAR ImagesIEEE Geoscience and Remote Sensing Letters, 2005
- A simple unsupervised MRF model based image segmentation approachIEEE Transactions on Image Processing, 2000
- Multiresolution Gauss-Markov random field models for texture segmentationIEEE Transactions on Image Processing, 1997
- An integration scheme for image segmentation and labeling based on Markov random field modelIEEE Transactions on Pattern Analysis and Machine Intelligence, 1996
- Improved fractal geometry based texture segmentation techniqueIEE Proceedings E Computers and Digital Techniques, 1993
- A review on image segmentation techniquesPattern Recognition, 1993
- Unsupervised texture segmentation using Markov random field modelsIEEE Transactions on Pattern Analysis and Machine Intelligence, 1991
- Stochastic and deterministic networks for texture segmentationIEEE Transactions on Acoustics, Speech, and Signal Processing, 1990
- A survey of thresholding techniquesComputer Vision, Graphics, and Image Processing, 1988
- Low-Level Segmentation of Aerial Images with Fuzzy ClusteringIEEE Transactions on Systems, Man, and Cybernetics, 1986