欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 计算机视觉 • 上一篇    下一篇

对称不变形状上下文特征提取与匹配

  

  1. 北京邮电大学信息与通信工程学院，北京 100876
  • 出版日期:2018-06-30 发布日期:2018-07-10
  • 基金资助:
    国家自然科学基金项目(61163044)；北京市科委基金课题(D171100003717003)；财政部项目(GSSKS-2015-035)

Symmetry Invariance Shape Context Feature Extraction and Image Matching

  1. School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • Online:2018-06-30 Published:2018-07-10

PDF (PC)

213

摘要： 形状上下文是一种广泛应用的图像形状特征提取与匹配算法，针对其特征不具有对
称不变性，无法对互相对称的相似图像建立匹配的问题，提出了一种具有对称不变性的改进形状
上下文特征提取与匹配算法。在形状边缘采样点上计算形状上下文中的角度关系描述时，令该点
的梯度方向为极坐标系的0°，并比较特征0 到π 与π 到2π 两个角度区间内其他边缘点的数量大
小，根据比较结果，调整极坐标系中角度增加的方向，从而使特征具备对称不变性。在迭代变形
与计算形状上下文时，仅在第一次迭代中使用改进的形状上下文特征，从而使匹配更加稳定。仿
真实验证明，该算法能够有效地在互相对称的相似图像间建立匹配，提高检索精度。

关键词: 图像匹配, 形状上下文, 对称不变性, 匹配稳定性

Abstract: Shape context is a popular algorithm of image shape feature extraction and matching,
traditional shape context algorithm lacks symmetry invariance, which means it can not match two
similar but symmetrical shapes. To overcome this shortcoming, this article proposed an improved
shape context algorithm with symmetry invariance. The method defines the gradient direction of a
sample point as its 0 rad in log polar coordinates when computing the angle relations in shape context,
then compare the number of sample points in 0 rad to π rad and π rad to 2π rad, and adjust the angel
increasing direction in the coordinate according to the results. In the process of iterative shape
warping and shape contexts calculation, the symmetry invariance descriptors are only applied in the
first iteration to obtain a more stable matching result. Experimental results show that the proposed
algorithm can effectively establish match between two similar and symmetrical shapes, and therefor
improve the retrieval precision.

Key words: image matching, shape context, symmetry invariance, matching stability