3D CAD model retrieval based on the combination of features

To improve the retrieval performance of 3D CAD model, we propose three combined feature descriptor and a class information based feature descriptor. First, we propose a novel feature descriptor Normal Angle Area (NAA) based on the 3D models surface and normals, and then we combined NAA with D2 and Bounding Box feature to form a more powerful feature DBNAA. Second, we combine DBNAA with Depth buffer images (DE), Ray extent (RE) to form DBNAA_DERE descriptor, and combine DBNAA with PANORAMA to form DBNAA_PANORAMA descriptor. Finally, we utilize the class information of dataset and propose a method called DBNAA_CBR which use a novel method to fuse class information into retrieval process. Experimental results on ESB dataset show that the results of DBNAA is similar to light field descriptors (LF) but with lower computational cost, DBNAA_DERE is better than DESIRE, DBNAA_PANORAMA performed better than all other non class information based methods. Further experiments showed that when we set an appropriate parameter for DBNAA_CBR it performs better than CBR-ZFDR.