Fast interactive segmentation using color and textural information

This work presents an efficient method for image multi-zone segmentation under human supervision. As most of the segmentation methods, our procedure relies on an energy minimization and this energy contains a data-consistent term and a coherence term. In this work, the pixel-by-pixel data-consistent term is based on a Support Vector Machine designed in order to deal with color and textural information. For the coherence term, our contribution is to take into account the gradient information of the image through the use of a Canny edge detector. To minimize the energy, as in Blake et al. ([1]), we use a Graph-Cut algorithm. Finally, we obtain excellent segmentation results at low computational cost. We compare our results on an image data bank on-line1. 1 http://research.microsoft.com/vision/cambridge/i3l/segmentation/ GrabCut.htm