Semantic-Oriented 3D Model Classification and Retrieval Using Gaussian Processes

The need of retrieving 3D models is constantly emerging. To improve the performance of a shape-based 3D model retrieval system, an approach is introduced to classify and retrieve 3D model by integrating shape features and semantic information. First, a new type of shape feature based on 2D views (called ZA) is proposed. Then we use Gaussian processes as supervised learning to mode the mapping from low-level features to query concepts. At last the method ranks models by the overall distance determined by a weighted sum of the semantic distance and the shape feature distance. Experimental results show that the performance of the 3D model’s multiclass classifier using proposed method is significantly higher than those of other supervised learning methods, and the retrieval can capture the query model’s high-level semantics, the retrieval performance is improved.