【摘要】 提出了一种新的轮廓图像检索算法。首先,根据内角大小将轮廓像素划分为凸角点、凹角点及平滑点3种类型,并分别提取其距离直方图;在此基础上,给出了两个新的轮廓空间特征描述符:相对位置分布及相关单元熵,分别描述角点间及平滑点间的相互关系;最后,结合上述3种特征,给出了轮廓间的相似性度量准则。实验结果表明,该算法取得了比传统方法更好的检索效果。更多还原

【Abstract】 The paper presents a novel approach for contour-based shape retrieval.The contour is firstly sampled and smoothed by a Gaussian filter.Then,the sampled points are classified into salient(convex or concave)or smooth type points by their internal angles.In addition to distance histograms(DH),two new descriptors named relative address distribution(RAD)and relative unit entropy(RUE)are introduced to describe the correlation of each type of contour points.These descriptors have powerful descriptive ability for contour with more spatial information.Comparisons are conducted between the proposed method and several other feature descriptors.The results show that the new method is efficient and it provides noticeable improvement to the performance of shape retrieval.更多还原