Embedded accelerated Gaussian model in graph cuts for automatic hand segmentation

Hand segmentation plays a great role in various computer vision areas, such as human computer interactive, sign language recognition and animation. In this paper, we propose a new method via accelerated Gaussian model (AGM) and graph cuts for automatic hand segmentation. The process consists of three stages, firstly, skin/non-skin seeds as hard constrains are generated based on AGM which is much faster than traditional Gaussian model (TGM) in our experimental results. Secondly, data term and smoothness term as soft constrains are defined in detail. Finally, graph cuts find the globally optimal segmentation by hard constrains and soft constrains. Comparative results demonstrate that our proposed method is more effective and robust than the well-known interactive algorithm, what´s more, it can automatically process hand segmentation without manual operation.