Evidential based Expectation Maximization for Image Segmentation

In statistics, expectation-maximization (EM) algorithm is an iterative method which finds the maximum likelihood or maximum a posteriori (MAP) estimates of parameters in statistical models depending on unobserved latent variables. In image segmentation, EM is widely used to determine the unknown parameters of different visual objects existing in an image. However, the main drawback of the EM method is that it does not consider spatial contextual information, which may entail rather noisy segmentation results. To remedy this, we develop an evidence theory based EM method (EEM) which incorporates spatial contextual information in EM by iteratively fusing the belief assignments of neighboring pixels to the central pixel. A simulated image set is used to evaluate the proposed method. Experimental results show that the new evidential method can achieve relative high accuracy.