Abstract: Shape context feature is not rotation invariant. Thus 2-D or 1-D Fourier transformation has been performed on the feature as improvement. But the improvement will increase the time complexity of feature extraction and matching. Based on the property of Fourier transformation, a fast and efficient method is presented in the cost matrix computation of these improved shape context features. Shape point matching experiment shows the efficiency and effectiveness of the new algorithm.

Key words: shape context feature, Fourier transformation, cost matrix

摘要： 形状上下文特征的最大缺憾在于不具备旋转不变性。为了解决这个问题可以采用二维或一维傅里叶变换来解决，然而这样改进会增加特征提取和匹配的运算时间。鉴于傅里叶变换的对称性，对于改进后的形状上下文特征在特征匹配步骤，提出代价矩阵的快速算法，能够减少运算时间。形状点集匹配的实验结果验证了这种新算法的有效性。

关键词: 形状上下文特征, 傅里叶变换, 代价矩阵