摘要：A new Student's t-distribution finite mixture model is proposedwhich incorporates the local spatial information of the pixels.The pixels' label probability proportions are explicitly modelledas probability vectors in the proposed model. We use the gradientdescend method to estimate the parameters of the proposed model.Comprehensive experiments are performed for synthetic and naturalgrayscale images. The experimental results demonstrate that thesuperiority of the proposed model over some other models.
