Hierarchical frame structure based interactive video object cutout

Video object cutout aims to extract homogenous objects from background in a video clip, which is a key process in many video processing fields, such as video compositing, video stylized rendering and so on. In this paper, we present a novel video cutout method by matching hierarchical structure of video frames. We first segment each frame by mean shift and construct hierarchical structure as a tree in preprocess stage. We then require user's interaction to label objects in a key frame. We further model video segmentation as matching hierarchical structure of frame and proposed an inter-frame matching algorithm. Experimental results show that our method can achieve desirable video segmentation results.

[1]  David Salesin,et al.  Video matting of complex scenes , 2002, SIGGRAPH.

[2]  Bo Thiesson,et al.  Image and Video Segmentation by Anisotropic Kernel Mean Shift , 2004, ECCV.

[3]  Daniel DeMenthon,et al.  SPATIO-TEMPORAL SEGMENTATION OF VIDEO BY HIERARCHICAL MEAN SHIFT ANALYSIS , 2002 .

[4]  Maneesh Agrawala,et al.  Interactive video cutout , 2005, ACM Trans. Graph..

[5]  Harry Shum,et al.  Video object cut and paste , 2005, ACM Trans. Graph..

[6]  Frédo Durand,et al.  A Topological Approach to Hierarchical Segmentation using Mean Shift , 2007, 2007 IEEE Conference on Computer Vision and Pattern Recognition.

[7]  Harry Shum,et al.  Video tooning , 2004, ACM Trans. Graph..

[8]  Guillermo Sapiro,et al.  Video SnapCut: robust video object cutout using localized classifiers , 2009, SIGGRAPH 2009.

[9]  Dorin Comaniciu Image segmentation using clustering with saddle point detection , 2002, Proceedings. International Conference on Image Processing.

[10]  Dorin Comaniciu,et al.  Mean Shift: A Robust Approach Toward Feature Space Analysis , 2002, IEEE Trans. Pattern Anal. Mach. Intell..

[11]  John P. Collomosse,et al.  Stroke surfaces: temporally coherent artistic animations from video , 2005, IEEE Transactions on Visualization and Computer Graphics.