New MRF Parameter Estimation Technique for Texture Image Segmentation using Hierarchical GMRF Model Based on Random Spatial Interaction and Mean Field Theory

This paper presents a new Markov random field (MRF) parameter estimation technique using hierarchical MRF model based on the random spatial interaction (RSI) and the mean field theory for the textured image segmentation. By considering spatial interaction of the MRF as random fields, the fluctuation of the spatial interaction that occurs in the conventional MRF model can be efficiently alleviated. Also, by assuming randomness of the spatial interaction as the MRF model, it allows us to obtain more robust information for segmentation during the feature extraction. The Gaussian MRF model is applied to the proposed hierarchical MRF scheme, and the expectation of the RSI is uniquely obtained by simple linear equation without using a window based on the mean field theory. Experimental results on synthetic and real world images show that the proposed algorithm provides good feature extraction and segmentation