30 likes | 199 Views
Statistical Techniques for Unsupervised Segmentation and Classification. TCD Interests Simon Wilson. Work so far. MRF-based greyscale image segmentation of textured images; IEEE Trans. Sig Proc, 2002, vol. 50, no. 2, pp. 357-365; Both labels and textures are modelled by MRFs (double MRF);
E N D
Statistical Techniques for Unsupervised Segmentation and Classification TCD Interests Simon Wilson
Work so far • MRF-based greyscale image segmentation of textured images; • IEEE Trans. Sig Proc, 2002, vol. 50, no. 2, pp. 357-365; • Both labels and textures are modelled by MRFs (double MRF); • Assumes no. of classes known, parameters of label and texture fields unknown; • Approach implemented with MCMC and simulated annealing to find MAP; • Applied to segmentation of satellite images; • Later work with INRIA-Ariana extended to case of unknown no. of classes – didn’t work so well – INRIA RR
Research Questions • These methods could be used in blind restoration/reconstruction with some appropriate modifications • Statistical methods of text recognition implemented through MCMC