Multi scale CRF based RGB-D image segmentation using inter frames potentials

This paper proposed a novel multi-scale approach to solve energy minimization problem which can be used to deal with indoor scene labeling problem. The principal idea of all multi-scale algorithms is solving their finer problems to efficiently initialize coarser. We propose the use of both color and depth information which is captured by Microsoft Kinect sensor. In order to create our energy function, we use the Conditional Random Field (CRF) approach, and add our geometrical constraint to pairwise potential as regions extraction method based on both edge of RGB and Range image and definition of Cliques between two consecutive frames. We evaluate our method on challenging NYU v1 dataset and Experimental results show that our proposed method reached 2.35 for Hausdorff criterion and enhances the time of image segmentation.