3d object retrieval approach based on directed acyclic graph lightfield feature

A new 3D object retrieval methodology is proposed by exploiting a novel directed acyclic graph lightfield feature (DLF), in which the so-called lightfield is made up of around 2000-5000 colour views. The descriptor overcomes the disadvantages of the existing view-based 3D object retrieval methods. Benefiting from shock graph and Bayesian network learning, this DLF is simple, accurate and noise robust as compared to existing methods. Results of experiments show that the proposed method is superior to others.