MRF parameter estimation based on weighted Least Squares Fit method

In general, the Markov random field (MRF) model parameter estimation is the first step in the image modeling applications. The accuracy of the parameter estimation will have a direct impact on whether or not the application based on the model can get the right result. In this paper, it is proposed a new method called the weighted least square fit, and then used the method in the MRF model parameter estimation. The method is based on the least squares fit (LS) method, and do some improvements so as could solve the LS method is not accurate enough and noise sensitive these disadvantages. And this paper have established an evaluation system based on the Expectation Maximization theory to evaluate the accuracy of the results generated by the different methods. Experiments have proved that if choose the same equations that selected from the field and under the same calculation complexity condition, the results based on this paper's method have a higher accuracy than the original LS method.