Prior knowledge-based feature selection for 3-D shape retrieval

Content-based 3-D shape retrieval is used in many fields. However, different feature vectors are used to capture the diverse characteristics of 3-D objects and no single feature vector is always able to outperform the others. This paper presents a prior knowledge-based dynamic feature selection algorithm for 3-D shape retrieval. The query model first calculates the prior knowledge of the feature vectors and then dynamically chooses the feature vector with the best description. Tests using the publicly available Princeton shape benchmark (PSB) 3-D model database show that the method significantly improves the retrieval effectiveness, with better retrieval performance than two state-of-the-art algorithms.