基于MDS的统计形状聚类

Statistical Shape Classification Based on MDS

针对物体轮廓曲线,采用新的模与相位保持的傅里叶描述方法,将连续的轮廓曲线降维归一化为RM空间中的点,引入DTW(Dynamic Time Warping)相似度,使用多维尺度分析(MDS),对多类物体基于形状进行聚类。实验结果表明,在合适的相似度下,该方法能够对不同的形状序列进行聚类。

A new normalized Fourier descriptor based on objects contour is defined, which keeps the amplitude and phase information of Fourier coefficients. Finding meaningful low - dimensional embedding in a high - dimensional shape space. Dynamic time warping （DTW） distance is used to measure the degree of similarity between two shapes. Shape classification is based on multidimensional scaling. Experimental results indicate that the approach provides satisfied classification.