中文摘要：

提出了一种新的基于轮廓的形状描述子,称为多尺度三元组描述子.对轮廓进行均匀采样,同时根据多边形近似演化算法提取轮廓的关键点,由采样点和其相邻关键点构成三元组,根据多个尺度下三元组的几何特性（包括角度和边长）定义描述子.这些三元组既包含了形状的局部细节,又包含了形状的全局结构信息,是一种稳定而准确的描述.形状匹配阶段使用动态规划算法.将本方法应用在MPEG-7数据库上,检索准确率达到86.30%,具有显著优势.

英文摘要：

Novel contour-based shape descriptor, multi-scale triple descriptor, was proposed. The original shape contour was sampled uniformly, and it was evolved by polygon approximation to get critical points. For every sample point, some triplets were defined using this sample point and its adjacent critical points. The sample point was described by the geometric measurement （including rotational angle and side lengths） of corresponding triplets in different scale levels. The generated triplets investigated the local details and global structure of shapes, which achieved stable and precise shape description. The dynamic programming technique was employed to perform shape matching. The novel descriptor achieved a retrieval rate of 86.30% on the MPEG-7 dataset, which significantly outperforms other methods.