针对变形体由于形状的变化,其不变量的构造在模式识别领域较困难等问题,对等距变形体的识别问题进行研究,提出一种等距变形体的矩不变量构造方法.首先利用三角网格上的快速行进算法来构造特征矩阵,使得点对之间的测地距离得以保存;然后,对特征矩阵进行归一化,保证同一目标特征矩阵的不变性;最后,构造矩不变量对归一化特征矩阵进行特征提取,并对该矩不变量的平移、尺度和旋转不变性进行证明.研究结果表明:与传统方法相比,该类不变量在不降低识别效果的前提下,运算复杂度较低,并对噪声具有较强的鲁棒性.%Based on the fact that the construction of invariants for isometric deformed object is difficult to research the pattern recognition, a new construction method of moment invariants for isometric deformed object was proposed to study the isometric deformed object recognition problem. First, fast marching method on triangulate domains was employed to compute the signature matrix to store the geodesic distance of every point pair. Secondly, the invariance was guaranteed by the normalization procedure on the signature matrices. Then, the moment invariants were constructed based on the normalized signature matrix, and the invariance of translation, scale and rotation were proved. The results show that the proposed method has less calculation complexity and stronger good noise robustness compared to traditional approach without reducing the recognition rate.
展开▼