3D augmented Markov random field for object recognition

In this paper, we propose to use 3D information to augment the Markov random field (MRF) model for object recognition. Conventional MRF for image-based object recognition usually uses appearance and 2D location as features in the model. The problem is solved by finding the globally optimal assignment that minimizes an energy defined in MRF. We estimate rough 3D information from stereo image pairs, and incorporate such information into node and edge potential models in the conventional MRF. Introducing 3D location into the node potential can take advantage of the 3D location distribution statistics of different classes. Considering 3D distance in the edge potential can help distinguish "true" neighbors from "fake" neighbors in 2D. Experiments show improved recognition results by using the proposed technique.

• Type: Journal
• Year: 2010
• Where: ICIP
• Authors: Wei Yu, Ahmed Bilal Ashraf, Yao-Jen Chang, Congcong Li, Tsuhan Chen
• Keywords: Conventional Mrf, Globally Optimal Assignment, ICIP 2010, Image Processing, Image-based Object Recognition