A Probabilistic Model for Traditional Chinese Architecture

Increasing numbers of 3D models provide a great opportunity for data-driven shape modeling, analysis and synthesis. The most critical core technique is to aggregate information from model collections to improve reasoning their properties and relationships. In this paper, we propose a probabilistic model for traditional Chinese architectures, which encodes the semantic type and hierarchical relationships for their basic components. Firstly a probabilistic hierarchical graph is designed to represent the typical component structure of Chinese ancient buildings. Secondly, a Bayesian Network is trained from a collection of 3D models with consistent labels. Finally, the Bayesian Network with structure and parameters learned from data can be used to synthesis and recommend components in applications. Experimental results show the effective of the proposed method.