以多尺度三角形为特征的快速形状匹配

摘要

为了减少形状匹配阶段的计算量并保持匹配精度,提出一种结合多尺度三角形特征的形状描述子,利用形状整体特征进行粗匹配并使用实时匹配方法完成匹配过程.首先通过高斯演化获得多尺度下的形状轮廓并用不同尺度上的轮廓点组成特征三角形;然后采用三角形面积和特征角来描述轮廓信息;最后用形状面积率对候选轮廓进行筛选,通过计算不同形状间描述子的相似度进行匹配.在不同形状数据集上进行实验的结果表明,该方法有较好的匹配结果和较少的匹配耗时,能够快速准确地描述形状的局部及整体特征,在实际应用中具有优势.%In order to obtain good accuracy and efficiency in shape matching, this paper presents a shape descrip-tor using multiscale triangle features. The descriptor is combined with Real-time Matching algorithm after a coarse-level matching using global shape feature. Firstly, Gaussian filter is applied to shape contours to achieve multiscale information. The method uses contour points on different scales to form triangles. Then it chooses the area and one angle of triangle as shape description. After filtering shapes by area rate, shape matching can be re-solved by calculating feature similarity between shapes. The method achieves good retrieval performance and ef-ficiency on several shape databases. The results show that the method can describe both local and global shape features well with low time complexity, and it has advantage on practical shape matching applications.