z轴旋转不变的三维形状描述子

z-axis rotational invariant 3D shape descriptor

针对现有三维模型内容特征提取过程中很难实现精准的预处理这一问题,提出了一种具有z轴旋转不变性质的三维形状描述子.此描述子基于模型体素化、球采样和傅里叶变换.在将模型体素化后,对体素模型进行多个同心球面采样,最后对采样数据应用三维傅里叶变换并提取其低频系数后形成模型的形状特征.在标准通用三维模型测试库Princeton Shape Benchmark上的实验结果表明:对于通用三维模型库,在无需模型位姿标准化的情况下,z轴旋转不变的三维形状描述子即可取得较好的检索效果,其表现优于形状分布、球面调和分析和直接傅里叶变换3种特征提取方法.

To solve the difficulty in precisely preprocessing a 3D model before the feature extraction process,a z-axis rotational invariant 3D shape descriptor was proposed which is based on voxelization,spherical sampling,and 3D Fourier transformation.A 3D model was first voxelized,and then a dataset from several concentric spherical samplings was used as the input for the 3D discrete Fourier transformation,while the absolute values of obtained coefficients were considered as components of the feature vector.The proposed approach was evaluated by the Princeton Shape Benchmark.Results show that a satisfactory result was achieved using the proposed approach.In comparison with feature extraction methods,such as shape distribution,spherical harmonics,and Fourier transformation,the z-axis rotational invariant 3D shape descriptor achieves better results in terms of retrieval effectiveness without normalizing the pose.