Abstract: For using descriptive information of individual feature of 3D models effectively,
the maximum similar category of the query model is determined by comprehensively analyzing
the distribution of models in retrieved results based on various features. Then the location entropy
of the same class models of individual feature is calculated respectively. Finally, combination
weights are computed dynamically based on the number of models of the maximum similar
category and location entropy. Compared with other dynamic and static combination methods on
Princeton 3D benchmark models, the results show the proposed method is effective in the case of
weak features included.

Key words: 3D model retrieval, multi-features, dynamic combination