Geodesic Star Convexity for interactive image segmentation Varun

Geodesic Star Convexity for interactive image segmentation Varun Gulshan , Carsten Rother , Antonio Criminisi , Andrew Blake and Andrew Zisserman † ‡ ‡ † Visual Geometry Group, University of Oxford, UK 1. Star-convexity 2. 1 Multiple Stars ‡ † ‡ Microsoft Research, Cambridge, UK 2. Star-convexity -- Extensions 2. 2 Geodesic Stars 3. Visibility Experiment Let y denote a binary segmentation and S*({c}) the set of star convex shapes wrt to center c. Star-convex constraint expressed as an energy: c q p Factorization into pairwise terms: Semantics I – visibility to atleast one center (equivalently: union of star shape sets) • The union constraint is not submodular. • Semantics don’t extend easily to brush strokes as star centers • Energy is submodular as (1, 0) labeling has infinite energy. • Only needs to be imposed for neighbouring pixels – hence efficient. c Geodesic Euclidean Veksler, ECCV 08 [1] Occlusion rates Method 1 star Semantics II – visibility to nearest star center • Tractable to implement (submodular) • Semantics extend nicely to brush strokes as star centers Visualization of geodesics computed using image gradients 4. Star-convexity in an interactive system 2 stars ESC 4. 50± 0. 58 2. 85± 0. 39 GSC 4. 16± 0. 54 2. 22± 0. 30 5. Evaluation BG stroke Boykov Jolly Energy: Simulated user [2] FG stroke Original image BG stroke User interaction Color likelihood BJ New Brush Stroke User interaction FG stroke Original image . . GSC Dij-GC [5] Dataset: 151 Images taken from Grab. Cut, PASCAL VOC and the alpha matting dataset. Evaluation criteria: Measure avg. number of strokes to reach desired accuracy. 4. 1 Sequential system Code and dataset available at: http: //www. robots. ox. ac. uk/~vgg/research/iseg/ Method Avg. Effort Output - BJ Output - GSC BJ 12. 35 SP 15. 14 PP 10. 66 BJ 12. 35 ESC 10. 57 RW 12. 31 GSC 10. 23 GSCseq 9. 63 SP = Shortest Paths [4] RW = Random Walker [3] PP = Post-Processing for connected components Comparison: Various shape constraints References Method Avg. Effort [1] O. Veksler. Star shape prior for graph-cut based image segmentation. ECCV, 2008 [2] H. Nickisch, P. Kohli and C. Rother. Learning an interactive segmentation system. ar. Xiv Technical Report, Dec. 2009. [3] L. Grady. Random walks for image segmentation. IEEE PAMI, 2006 Comparison: Different algorithms [4] X. Bai and G. Sapiro. Geodesic matting: A framework for fast interactive image and video segmentation and matting. IJCV 2009 [5] S. Vicente, V. Kolmogorov and C. Rother. Graph cut based image segmentation with connectivity priors. CVPR 2008.