基于多尺度投影的相似颅骨检索

Similar Skull Retrieval from Multiscale Projection

在基于模板变形的颅面复原方法中,复原的开始阶段需要在数据库中选取与待复原颅骨最为相似的参考颅骨.鉴于基于三维模型的检索算法时间久且颅骨间的差异细微,从而不同于一般三维模型数据库中各模型的差异.因此,已有的三维模型检索算法不适用于颅骨检索.本文提出一种夹角信息和距离信息融合的颅骨轮廓特征提取算法,并在此基础上提出一种能够反映颅骨空域信息的剖面特征提取算法.检索时首先获取三维颅骨的正交投影和深度投影,通过正交投影获取轮廓的角度和距离特征,通过深度投影获得具有空域信息的剖面特征;然后对多个特征进行加权融合搜索到最相似颅骨;最后通过ICP+TPS对检索到的颅骨进行误差评估.实验表明,本算法在保证检索效率的同时,可以准确地应用在颅面复原前期对最相似颅骨的选择上.

In the template-based method of craniofacial reconstruction, the first step is to select the reference skull in the database which is the most similar to the skull being restored. 3D model retrieval algorithms usually take a long time, and the differences between skulls are so tiny that the general models in 3D model database cannot differentiate them, so any general 3D retrieval algorithm is not suitable for 3D skull retrieving. To cope with these problems, an algorithm for skull contour feature extraction with integration of angle and distance information is proposed in this paper. On the basis of this, a new profile feature extraction algorithm reflecting the spatial information of skull is also proposed. In the process of retrieving the most similar skull, firstly, orthogonal projection and depth projection of 3D skull are obtained. The angle and distance features of profiles are obtained by orthogonal projection while the profile features of spatial information are obtained by depth projection. Then, multiple features are fused according to their weights to search for the most similar skull. Finally, the error of the retrieved skull is evaluated by ICP ~ TPS algorithm. Experimental results show that the algorithm can ensure retrieval efficiency. Meanwhile, it can be accurately applied to craniofacial reconstruction of earlier stage to find out the most similar skull.