Superpixel based RGB-D image segmentation using Markov random field
Taha Hamedani, Ramin Zarei, Ahad Harati · 2015
In this work we proposed a novel super pixel based segmentation approach to solve energy minimization problem which can be used to deal with indoor scene labeling problem. We used Range data beside color image captured from Kinect sensor. This sensor enables us to use 3D features of structure like normal vector and 2D color features. We extracted the region of scene as super pixel based on the both color and direction change; and, consequently, we constructed our graphical model on these regions and apply Markov random field inference to assign efficient labels to them. Our evaluation on 30 scenes of challenging NYU v1 dataset shows that our proposed method reached higher values of “Correct Detection” and lower rate of “Missed instances” and “Noise instances” criteria according to Hoover evaluation method.