Skip to Main Content
In this paper, we propose an efficient approach to automatic human object segmentation. First, foreground (human object) model and background model are built based on the face detection result, and are used to obtain the seed pixels for foreground and background, respectively. Then seed pixels are clustered using K-means algorithm, and Gaussian mixture models are exploited to generate the foreground/background probability map. Finally, pixels are efficiently classified into foreground and background under the framework of graph cuts. Experimental results on a variety of video sequences demonstrate the better segmentation performance of the proposed approach.
Date of Conference: 23-25 Nov. 2010