Interactive Object Selection Using S-t Minimum Cut

In this paper, we present an interactive object selection approach that applies s − t minimum cut on local contour neighborhoods of the object boundary to segment the object from the background. Using a simple interface, users can extract accurate object contours interactively. Our approach has the following properties. 1) Interactivity: The interface is easy to use and accepts flexible freehand inputs. Realtime feedback is provided for user interactive correction. 2) Robustness: The s − t minimum cut is insensitive to image noise since it finds the global minimum within the local contour neighborhood. Further, the approach is valid for contours and objects defined by different features, e.g., color and texture. 3) Controllability: Object contours are fully controllable through interactive correction by the user. Experimental results for a wide range of objects are provided.