3D CAD Model Clustering and Retrieval Inspired by Fish Swarm

In order to improve both the discriminative power for local regions of CAD model and the searching efficiency in 3D CAD model retrieval, a novel method for 3D CAD model clustering and retrieval is proposed inspired by fish swarm. According to the Attribute Adjacent Graph（AAG） of B-rep model and spectral graph theory, each CAD in library model is represented by a shape descriptor called as CAD Spatial Bag of Words（CMSBows）, which is combined with the adjacent relations of the local regions of CAD model. An improved Artificial Fish Swarm Algorithm（AFSA）, based on the global bulletin board guiding and the fuzzy c means revising, is proposed to partition the global database into different model base. For model retrieval, a two-level searching mechanism is employed, i.e., the query model is classified into the corresponding sub model base using membership function, and then, the most similarity CAD model is retrieved in this limited scope. Experimental results show that the proposed feature descriptors improve the distinguishing ability for model’s local parts, and the clustering effectiveness of the proposed methods is good. The methods improve both retrieval quality and efficiency significantly, so it can achieve the effective reuse of CAD model.