Abstract: Aiming at the problem that the important points of the contour curve may be omitted with shape signature representation method, this paper proposes a centroid distance signature method based on corner point detection. It uses contour tracing to get the contour, adopts corner point detection to obtain corner points, and represents shape signature with centroid distance. Experiments on 1-NN classification and K-medoids clustering show the effectiveness of the method.

Key words: shape feature, feature extraction, shape signature, time series, corner point

摘要： 针对形状标记表示法可能遗漏轮廓曲线上较重要边界点的问题，提出一种基于角点检测的质心距离标记法。该方法采用轮廓跟踪技术获取物体的轮廓曲线，通过角点检测得到曲线上的所有角点，利用质心距离产生形状标记。以最近邻分类和K-中心聚类对其性能进行验证，结果证明了该方法的有效性。

关键词: 形状特征, 特征提取, 形状标记, 时间序列, 角点

CLC Number: 

• TN911.73