An improved GrabCut on multiscale features

Abstract

the GrabCut can effectively extract the foreground according to features in a cartoon image; however, the performance is not so effective for a real image, because the feature extraction is independent of segmentation. To improve segmentation performance, this paper proposes an improved GrabCut which combines the segmentation and multiscale feature extraction into a unified model. In this model, the segmentation relies on multiscale features, and the multiscale features depend on multiscale decomposition. A novel total variation regularization is proposed in multiscale decomposition to preserve edges and remove the region inhomogeneity, by which the generalization of features for segmentation is improved. The features obtained by the multiscale decomposition are integrated into the segmentation process, and the foreground can be easily extracted from a proper scale. Experimental results indicate that, compared to the existing GrabCut and improved techniques, this method provides competitive performance in terms of the segmentation accuracy, while being insensitive to inhomogeneity.