Automatic people segmentation with a template-driven graph cut

This paper presents a new fully automatic method for segmenting upright people in the images. Is is based on the efficient graph cut segmentation. Since colour and texture prevent from discriminating this particular class, silhouette shape is used instead. The graph cut is guided by a non-binary template of silhouette that represents the probability of each pixel to be a part of the person to segment. Subsequently, part-based template is used to better take into account the different postures of a person. Our method is close to real time and is tested on a large person dataset.