In the process of shape matching based on shape context, each of the samples is directly classified into one bin in the histogram, which leads to the inaccurate feature representation and matching bias results. With fuzzy membership function introduced into log-polar coordinate, fuzzy shape context is constructed based on the histogram which represents the fuzzy partition on the distribution of samples, and shapes can be accurately represented by the descriptor. The sample set is divided into subsets in polar coordinates, and segment matching is proposed for shape matching with incorrect matches eliminated. On this basis, the circular shift matching is adopted for resolving the problem of shape matching in different postures. Experimental results prove that shape recognition and retrieval can be effectively achieved by using the proposed method.%利用形状上下文特征进行形状匹配的过程中,各采样点被直接二值划分至不同的直方图栅格,致使特征表达不精确,进而导致匹配结果存在偏差.本文在对数极坐标系中引入模糊隶属度函数,利用采样点分布的模糊划分结果建立直方图,生成模糊形状上下文特征,从而更精确地描述形状信息.在极坐标系下对采样点集合进行分割,提出分割匹配的方法,减少不必要的特征匹配次数.在此基础上,利用循环移位匹配方法解决形状在不同角度姿态下利用形状上下文特征匹配的问题.通过对不同数据进行仿真分析,证明本文所提出的方法能有效实现形状识别和检索.
展开▼
