Real-timeVideoSegmentation Using Student'stMixture Model

Mixture models for video segmentation have mainly revolved around Gaussian distributions for a long time due to their simplicity and applicability. In this work, we propose a novel real-time video segmentation algorithm based on Student’s t mixture model. Though, Student’s t-distribution has been used for image segmentation by applying Expectation Maximization (EM) algorithm, the same technique cannot be followed in video segmentation due to exceptional increase in computational complexity. Thus, in spite of being a more heavily-tailed distribution compared to Gaussian, Student’s t mixture model remained unexplored for video segmentation. In this work, a novel and effective recursive filter based formulation has been introduced to update the mixture model with new observations. Our analysis and experimental results show that real-time, robust and improved video segmentation can be performed using Student’s t mixture model compared to the conventional mixture models. c © 2012 Published by Elsevier Ltd.