Video Object Segmentation with a Potts Model
Jieyu Zhao, Xiaoquan Wang · 2007
This paper presents a probabilistic graphical model, a Potts model with external fields, to solve a challenging video object segmentation problem. The video image is represented with a weighted graph. A dynamic Potts model with external fields is used to store the probability distribution of the image pixels. The external fields are important for weighting terms in mixture distributions and thus allow more flexible and robust image segmentation. An online expectation-maximization (EM) algorithm is developed to estimate the parameters of the model. Experimental results on different video clips show the proposed approach is capable of retrieving different objects such as cars, planes, animals and human beings effectively.