【作者】 秦实宏； 韩水华；

【Author】 Qin Shihong Han Shuihua Huazhong University of Science and Technology, Wuhan, 430074

【机构】 华中科技大学电力系； 华中科技大学计算机学院；

【摘要】 一个鲁棒的形状匹配算法不仅要具备对旋转、位移和比例的不变性,而且应保持对噪声、分割不一致引起的形状扭曲无关性。本文采用外轮廓曲线的形状描述方法,提出演化多边形形状匹配的新算法。利用Harr变换方法,通过设置不同的最大变换允许误差,不仅可以得到近似的轮廓多边形并且还构成一个形状的层次结构,采用多阶段的“粗粒度”和“细粒度”相结合的相似匹配方法,可实现快速有效的形状图像检索。更多还原

【Abstract】 A robust shape matching approach should be invariable to rotation transition and scale as well as independent to the distortion caused by noise and segmentation error. In this paper, a multiscale shape matching approach is presented based on discrete curve evolution. Using Harr transformation and given different maximum transformation error, it not only decomposes 2D object into polygonal curve but also induces a hierarchical structure of shape. Combing "Coarse level" with "fine level" in hierarchical similar matching, the retrieval of shape image can be effective for both retrieval accuracy and efficiency.更多还原