3D model retrieval by shape similarity

Abstract

The objective of this project is to develop a content-based 3D object model retrieval system by making use of the shape features such as distance, curvature and surface normal. Also, provide a convenient way for users to search similar 3D models in the database more efficiently. The use of surface normal is to shortlist the models from the database. This can reduce the computation time in the later stages. There are two methods for calculating distance. Method 1: Distance is the calculation of the length between a vertex of the object model and the object centre being measured. Method 2: First determine the number of sampling points and find coordinates of each sampling point on the surface of a unit sphere. Then, shoot imaginary rays from each sampling point to the centre of the unit sphere. Compute intersection points and calculate distances between the centre of the sphere and the intersection point on the object surface. Curvature is the measure of the vertex angle of the object model. To retrieve the most similar models, the query is compared with each instance of the database using these shape features. The VRML 2.0 File Format 3D models have been used as the testing data. These models are stored in a 3D model database with extracted shape features. After testing the program with the 3D models in database, I found that the rating for curvature distribution and distance distribution is a great factor that affects the retrieval result. Therefore, a suitable rating will give you a reasonable result. The detailed result will be described in the report.