An Efficient and Effective Image Segmentation Approach using Spatially Constrained Finite Mixture Model

This study introduces a novel image segmentation approach based on clustering using finite mixture model. The proposed approach utilizes the Kullback-Leibler divergence as the prior probability to incorporate spatial information into a mixture model. To alleviate the complicated learning process after incorporating the spatial prior, a multinomial logistic model is adapted, and a novel entropy approximation method is introduced so that the learning process keeps the same time complexity as in standard expectation maximization algorithm. Further, a noise removal and edge preserving method is proposed to deal with the under-smoothing and over-smoothing problem for better segmentation results. Experiments using synthetic and real images of Computer Tomography (CT) scan of the teeth are conducted, and the study shows convincing results obtained by the proposed approach.