Applications Of Mean Shift Algorithm On The Computer Vision

Mean shift is an effective iterative algorithm widely used in clustering, tracking, segmentation, discontinuity preserving smoothing, filtering, edge detection, and information fusion etc. This thesis introduces a non parametric procedure for the analysis of multimodal data based on mean shift property, and demonstrates its superior performance in various applications. The convergence of the mean shift procedure to the closest mode of the underlying distribution is proven, both for the Epanechnikov kernel and the general case of kernels with convex and monotonic decreasing profile. The smooth trajectory property of the mean shift is also demonstrated. Significant feature in the image correspond to high density regions in feature space. The number of clusters and the cluster center are automatically derived by mode seeking with the mean shift procedure. Through density limited and space limited,we can have a reduced set of points randomly selected from the data. The cluster boundaries are delineated using a k-nearest neighbor's technique. Many examples of color image segmentation using mean shift procedure have been showed in this thesis. The aim of image segmentation is to mark and locate targets and background in the image according to their respective prior knowledge, and then separate the targets to be recognized from background and other false targets. The segmentation processes in the joint image domain and feature space is robust and performs well. In this thesis, presents a fusion method that combing mean shift and particle filter. It also integrates the efficiency and robust excellence of the two methods. Finally experimental results show that the real-time and robust tracking of the fusion method in complex background.