Shortest path based planar graph cuts for Bi-layer segmentation of binocular stereo video

Abstract

Separating a foreground layer from stereo video in real-time is used in many applications such as live background substitution. Conventional separating models using stereo, contrast or color alone are usually not accurate enough to be satisfactory. Furthermore, the powerful tool of graph cut which is well suited for segmentation is known to be not efficient enough especially for high resolution images. In this paper, we conquer these difficulties by fusing stereo with color and contrast to model the segmentation problem as an minimum cut problem of a planar graph and solving it by a specialized algorithm, parametric shortest paths [8] with a dynamic tree structure, in O(nlogn) time. Experimental results demonstrate the high accuracy and efficiency of the algorithm. © 2011 Springer-Verlag Berlin Heidelberg.