Improved Cut-Based Foreground Identification

Automatic content based schemes, as opposed to those with human endeavor, have become important as users attempt to organize massive data presented in the form of multimedia data such as images, and home or movie videos. One important goal, be it in shot understanding, or scene detection, or compression, is the ability to find foreground pixels. This higher level task is best realized using a graphbased description of the input image or video. The normalized cut framework is appealing because it looks at an image or an image sequence from a global perspective. Unfortunately due to quadratic storage and time complexity, the algorithm appears to be infeasible to use on medium and large datasets. In this paper, we show how to make graph based schemes tractable and useful.