RGBD image segmentation

In this paper we present a method to segment RGBD image of a scene into coherent and meaningful parts using both the appearance features and depth information. The segmentation method is totally based on graph cuts theory which uses our proposed unsupervised Conditional Random Field (CRF) model. We evaluate our method both quantitatively and qualitatively on a set of RGBD images of NYU dataset. The results show that the combination of unsupervised CRF with graph cuts can be as accurate as supervised methods and in some cases can perform better than other segmentation methods.