融合特征描述符约束的3维等距模型对应关系计算

Calculation of correspondences between three-dimensional isometric shapes with the use of a fused feature descriptor

目的为了更准确地构建3维等距模型之间的对应关系,本文提出了一种基于热核签名与波核签名的融合特征描述符计算3维等距模型对应关系的方法。方法首先计算3维模型Laplace算子获得模型的特征向量和特征值;然后将所得到特征值和特征向量作为基参数分别计算源模型与目标模型的热核签名和波核签名,并将热核签名与波核签名融合为一个新的特征描述符。融合特征描述符作为模型上随机均匀采样点的约束,通过最小值匹配算法得到源模型和目标模型之间的对应关系。结果实验结果表明,利用融合特征描述符约束进行计算得到的对应关系正确匹配率比热核签名约束计算得到的对应关系匹配率平均提高19.429%,比波核签名约束计算得到的对应关系匹配率平均提高4.857%。结论本文提出的融合特征描述符适用于计算3维等距模型或近似等距的3维模型之间的对应关系,与单一使用热核签名或波核签名特征描述符相比,可以得到更加准确的对应关系。

Objective A new fused feature descriptor based on heat kernel and wave kernel signatures is proposed to calcu- late shape correspondences between isometric 3D models. Method First, the Laplace operators of 3D models are calculated to obtain eigenvectors and eigenvalues, which are defined as the basic parameters for computing the heat kernel and wave kernel signatures of the source and target models, respectively. Second, heat kernel and wave kernel signatures are fused together as a new feature descriptor, which serves as a constraint for points sampled uniformly in random. Finally, the shape correspondences between models are calculated by a minimum-value matching algorithm. Result Experimental results show that the accuracy of the correspondence ratio calculated using the proposed feature-descriptor constraint increases by 19. 429% and 4. 857% on average with respect to the correspondence ratios calculated using heat kernel signature and wave kernel signature constraints, respectively. Conclusion The fused feature descriptor presented in this article is applicable to calculating correspondences between isometric 3D models or approximately isometric 3D models. Compared with the descriptors that only use heat kernel or wave kernel signatures, the fused-feature descriptor was able to obtain more accurate correspondences.