【摘要】 利用等间隔方法在物体表面标定特征标记点构成训练样本集,所获得的统计形状模型质量较低。针对该问题,提出一种基于粒子滤波器算法的特征标记点匹配算法。通过球面保角映射将三维表面映射到二维变量化空间,并利用粒子滤波器算法框架将物体表面局部几何特征量和整体空间结构特征相结合,实现特征点的最优匹配。实验结果表明,与利用等间隔方法相比,该算法获得的统计形状模型具有更高的通用性和专一性。更多还原

【Abstract】 Groupwise surface correspondence is the crucial step to construct the Statistical Shape Model( SSM).How ever,the quality of the models using the equally-placed landmarks method is not sufficient. Aiming at this problem,this paper proposes a novel correspondence method using particle filte algorithm. All the 3D training surfaces are mapped to a unified spherical parameter space based on the spherical conformal mapping at first. The corresponding parts of the surfaces can be found according to the local geometric characters as well as the global structures,which can be integrated by the frameworks of particle filters. Experimental results illustrate that compared with the equally-placed landmarks method,the SSM models obtained by proposed method can perform well on the general ability and the specific ability.更多还原