Interactive Segmentation of High-Resolution Video Content Using Temporally Coherent Superpixels and Graph Cut

Interactive video segmentation has become a popular topic in computer vision and computer graphics. Discrete optimization using maximum ow algorithms is one of the preferred techniques to perform interactive video segmentation. This paper extends pixel based graph cut approaches to overcome the problem of high memory requirements. The basic idea is to use a graph cut optimization framework on top of temporally coherent superpixels. While grouping spatially coherent pixels sharing similar color, these algorithms additionally exploit the temporal connections between those image regions. Thereby the number of variables in the optimization framework is severely reduced. The e ectiveness of the proposed algorithm is shown quantitatively, qualitatively and through timing comparisons of di erent temporally coherent superpixel approaches. Experiments on video sequences show that temporally coherent superpixels lead to signi cant speed-up and reduced memory consumption. Thus, video sequences can be interactively segmented in a more e cient manner while producing better segmentation quality when compared to other approaches.