Abstract: A novel 3D object retrieval algorithm is proposed for meeting 3D object retrieval.Firstly，based on the shortcomings analysis of existing view-based 3D model descriptors，many 3D model descriptors can not adequately describe 3D object information.So the concept of lightfield hybrid descriptor（denoted as H_D） is developed.The H_D includes 3 sub-descriptors，i.e.color descriptor H_CD based on color histogram，texture descriptor H_TD based on Bayesian network learning and shape descriptor H_SD based on shock graph.Then overall framework of H_D is discussed，adaptive clustering algorithm is used to gain Characteristic View （CV），and then the method that how to generate H_D and the corresponding metric mechanism are described.In the end，some experiments are done to analyze retrieval performance based on H_D.The results show that the proposed method is superior to other view-based methods.

Key words: 3D object retrieval, Bayesian network, light field hybrid, descriptor

摘要： 针对三维目标（3D object）检索问题，提出了一种基于新型描述符的3D目标检索方法。首先，在分析现行基于视图的3D模型描述符在描述方法上不充分的基础上，提出了混合描述符H_D的总体思路。进而讨论了H_D总体框架，即在光场图像阵列自适应的基础上，实现了直方图颜色描述符H_CD，shock图形状描述符H_SD及贝叶斯网络（Bayesian Network，BN）纹理描述符H_TD的优化组合。其次，讨论了H_D各部分的具体实现及度量机制，最后，对H_D检索性能进行了实验分析，结果表明提出的方法是优于其他基于视图的检索方法。

关键词: 三维目标检索, 贝叶斯网络, 混合光场, 描述符