Abstract Intelligent Scissors and Intelligent Paint are complemen- tary interactive image segmentation tools that allow a user to quickly and accurately select objects of interest. This demonstration provides a means for participants to experi- ence the dynamic nature of these tools. 1. Introduction No fully automatic image segmentation technique cur- rently yeilds satisfactory results over a broad class of image sources, content, and complexity. Thus, intelligent segmen- tation tools which exploit high level vision methods while requiring minimal human interaction become appealing. Intelligent Scissors and Intelligent Paint are complementary, general-purpose, object selection tools which allow rapid and accurate object segmentation from arbitrarily complex backgrounds using simple gesture motions with a mouse. Intelligent Paint allows a user to efficiently select the region associated with an object of interest while with Intelligent Scissors, a human operator can quickly define object bound- aries. Both tools utilize tobogganing to partition an image into homogeneous regions which are effectively identical to the catchment basins produced by computing the watershed of the image’s gradient magnitude map. These regions become the basis for region-based (Intelligent Paint) or boundary-based (Intelligent Scissors) object selection. This demonstration provides hands-on experience with these two interactive object selection tools. Due to the dynamic nature of these tools’ input and feedback, it is diffi- cult to convey the utility of Intelligent Scissors and Intelli- gent Paint in print (or even in video). Unlike previous object selection methods where segmentation typically occurs only after user input is complete, Intelligent Scissors and Intelli- gent Paint integrate human input with the segmentation pro- cess to provide dynamic interaction and feedback between the user and the computation. Thus, this demonstration will allow participants to directly evaluate the usability and ben- efit of these two complementary tools. 2. Intelligent Scissors Intelligent Scissors is a general purpose, interactive object selection tool which allows a user to choose a mini- mum cost contour segment corresponding to a portion of the desired object boundary. As the mouse position comes in proximity to an object edge, a live-wire [1] boundary “snaps” to and wraps around the object of interest. Since the user interactively selects a boundary from a collection of optimal solutions, the user knows exactly what the resulting contour will be during interaction. Intelligent Scissors achieves this goal by imposing a weighted graph on the image and interactively computing the optimal path from a user selected “seed” point to all other points in the image using an efficient, linear-time implementation of Dijkstra’s graph search. As the cursor moves, the optimal path from the pointer position to the seed point is displayed, allowing the user to select an opti- mal contour segment which visually corresponds to a por- tion of the desired object boundary. Compared to manual tracing (i.e., lassoing), object selection using Intelligent Scissors is many times faster, more accurate, and more reproducible [2] (see Figures 1 and 2). Last year, we presented toboggan-based Intelligent Scis- sors [3] which uses tobogganing to partition an image into homogeneous regions and then imposes a weighted, planar graph onto the region boundaries. The region-based graph is many times smaller than the pixel-based version used in our previous Intelligent Scissors [2]. This reduces the size of the graph which greatly speeds up the graph search and increases interactive responsiveness. Further, the tobog- ganed regions provide an efficient framework for computing a 4-parameter edge model (providing sub-pixel localization, blur, and foreground/background color) and improved cur- sor snapping (to facilitate cursor placement on object edges). 3. Intelligent Paint Intelligent Paint [4] uses a simple connect-and-collect strategy to quickly and accurately define an object’s region. This strategy uses a new hierarchical tobogganing algorithm to automatically connect image regions that naturally flow together, and a user-guided, cumulative cost-ordered expan- sion interface to interactively collect those regions which constitute the object of interest. Each paint stroke interac- tively samples the properties of the underlying regions as the intelligent paint adaptively flows into similar neighbor- ing regions. This strategy coordinates human-computer interaction to extract complex regions of interest from arbi- trary backgrounds using simple paint strokes with a mouse. While tobogganing at the pixel resolution creates highly oversegmented regions, subsequent tobogganing between previously computed regions produces a natural hierarchy of segmentations. Hierarchical tobogganing produces seg- mentations which are not dependent on heuristic parameters or thresholds, and which are largely associated with image objects at multiple resolutions (see Fig. 3). Using Intelligent Paint is like using a can of spray paint—with a few enhancements. The user simply presses the mouse or pointer button to begin painting at the cursor position, then releases the button when finished painting (see Figures 4 and 5). While the button is pressed, Intelli- gent Paint does two things: 1) it automatically expands the painted region to fill the object of interest, and 2) it interac- tively responds to user input. Enhancements to this basic press-and-release interface include an undo feature to unpaint regions, a slow motion feature to paint with greater precision, and an Intelligent Eraser. Intelligent Selection Tools Eric N. Mortensen L. Jack Reese William A. Barrett Brigham Young University 1063-6919/00 $10.00 ã 2000 IEEE