Statistical Learning for Remote Sensing Image Segmentation

This project includes various research areas including mathematical theory, pattern recognition and random processes. Applications of this project include density estimation, image segmentation and classification.

         Research Team
Aly A. Farag CVIP Director farag@cvip.uofl.edu
Refaat M. Mohamed Research Assistant refaat@cvip.uofl.edu
Methods

This project uses statistical learning for the segmentation of remote sensing imagery. It contains learning of Support Vector Machines using Mean Field theory for Density Estimation in high dimensional data sets. A Bayesian classification method based on the estimation of the class probability density function using SVM is used. To incorporate the contextual information in the segmentation process, Markov Random Field  (MRF) modeling is used. An analytical approach for estimating the clique coefficients of Gibbs Markov Random Fields is used. Map refinement using Iterative Conditional Modes (ICM) algorithm and the analytical estimation of the clique coefficients of the GMRF are applied.

Results
RGB from a 34-band hyperspectral data set

Segmentation results using MF-based SVM algorithm

 Segmentation refinement using MRF modeling

Publications

Aly A. Farag, Refaat M. Mohamed and A. El-Baz, “A Unified Framework for MAP Estimation in Remote Sensing Image Segmentation,” IEEE Transactions on Geosciences and Remote Sensing, vol. 14, no. 8, pp. 1617-1634, July 2005.

Refaat M. Mohamed, A. El-Baz, and Aly A. Farag, “Probability Density Estimation Using Advanced Support Vector Machines and the Expectation Maximization Algorithm”, International Journal of Signal Processing, vol. 1, pp. 185-188, March, 2005.

A. El-Baz, Refaat M. Mohamed, and Aly A. Farag, “Advanced Support Vector Machines for Image Modeling Using Gibbs-Markov Random Field”, International Journal of Computational Intelligence, vol. 1, pp. 306-309, March, 2005.

A. Farag, A. El-Baz, and Refaat M. Mohamed, “Density estimation using generalized linear model and a linear combination of Gaussians, ” International Journal of Signal Processing, 2005, vol. 1, pp. 76-79, March, 2005.

Acknowledgement / Sponsors
We would like to thank the Air Force Office for Scientific Research.