Medical image segmentation using characteristic function of Gaussian Mixture Models

Gaussian Mixture Models (GMMs) have interesting properties that make them useful for many different image applications because they have powerful probabilistic statistical theory basis. However, the application of GMMs to medical image segmentation faces some difficulties. First, many typical model selection criterions become invalid when they estimate the number of components of medical images. Second, the convergence function of GMMs suffers slow convergence. In this paper, a novel medical image segmentation method based on characteristic function of GMMs is proposed. First, a new model selection criterion using characteristic function of GMMs is proposed to estimate the number of components in medical image. Second, a new convergence function using characteristic function of GMMs is proposed to estimate the parameters of GMMs. The experimental results of CT image segmentation show that our algorithm achieves better results than those from many derivatives of GMMs and needs less computation time.