典型时间规整算法支持下的建筑物形状相似性度量

A canonical time warping algorithm for building shape similarity measurement

本文提出了一种基于典型时间规整(canonical time warping,CTW)算法的形状相似性度量模型,该模型运用动态时间规整(dynamic time warping,DTW)算法和典型相关分析(canonical correlation analysis,CCA),对齐具有不同节点个数的建筑物坐标序列,并综合评价不同形状轮廓之间的相似性。该方法直接以矢量坐标作为模型输入而无须复杂的形状编码,顾及了建筑物图形原始的轮廓特征,可高效地应用于形状检索等场景。试验表明,CTW算法在度量形状相似性时具有平移、旋转、缩放和镜像不变性,能有效度量建筑物形状之间的形状相似性,其结果符合人类的空间视觉认知。

This paper proposes a shape similarity measurement model based on canonical time warping(CTW)algorithm.The model combines canonical correlation analysis(CCA)and dynamic time warping(DTW)to align building coordinate sequences with different number of vertices,which can comprehensively evaluate the shape similarity between different shape contours.This method directly uses vector coordinates as model input without constructing complex shape coding and considers the original contour features of building shapes,which can be applied efficiently to shape retrieval and other scenarios.Experiments show that CTW algorithm is invariant to translation,rotation,scaling and mirroring when used to measure the similarity of geometric objects,and can effectively measure the shape similarity between building shapes.The results are consistent with human spatial visual cognition.