3D model retrieval from multiple photographic images

Abstract

Due to the recent significant improvements in game entertainment, CAD, CAE and some other 3D fields, the number of 3D models is increasing rapidly. As a result, there is an increasing need for procedures supporting the automatic search for 3D objects in databases. In this paper, we present a novel framework to retrieve 3D models in 3 steps. Firstly, users input multiple ordered images of an object as input. Certain photo environment is needed. A voxel model will be generated automatically according to these inputs. Secondly, we provide a tool to modify voxel model. This procedure can help eliminate defects caused by uncertain photo environments. If the model produced by the first step is reasonable, this step will be optional. Finally, the modified model will be retrieved by a method called Solid-D2, which provides a fast and discriminating descriptor for 3D shapes. This algorithm is chosen to err on the side of retrieval speed with certain precision. Related models from the database will be listed. Experimental results indicate that the proposed method is easy-to-use, efficient and applicable for 3D model retrieval.