Tracking object contour based on multiple features
基于多特征的目标轮廓跟踪*

Abstract: In this paper, Kernel density estimate is used to construct the probability distributions models for color feature and texture feature. With these two models, Bayesian model calculates the posterior probabilities of the object and background pixels. Furthermore, a new region energy functional is proposed to count the energy of the object and the background pixels respectively.So that minimum energy contour coincides with the object contour. At last, the gradient descent flow derived from the variational reduces the contour energy and converge to the object boundary by evolving it. Different experimental results show that the proposed algorithm can track the rigid object contour and non-rigid object contour in image sequences efficiently.