光学 精密工程, 2020, 28 (4): 963, 网络出版: 2020-07-02
基于SURF特征描述符和杰卡德距离的文物碎片拼接
Mosaic of cultural relics fragments based on SURF feature extraction descriptor and Jaccard distance
碎片拼接 特征提取 SURF特征描述子 边缘提取 刚体变化 fragment splicing feature extraction Speeded Up Robust Features(SURF) descriptor edge extraction rigid body change
摘要
针对文物碎片拼接过程中存在因局部碎片缺失和纹饰几何特征受损导致断裂面特征提取不准确的问题, 本文提出了一种基于SURF特征描述符和杰卡德距离(Jaccard Distance)的文物碎片拼接方法。首先用Canny算子提取碎片边沿轮廓线以及模型表面纹饰线条, 再通过构造多尺度空间来提取断裂面特征点。其次, 针对以往构建特征描述符存在的高冗余和欧氏距离计算存在的高延时问题, 本文提出构造低冗余的SURF特征描述符, 再借助Jaccard距离来进行特征点的相似性比较, 确定碎片最优邻接关系。最后采用ICP方法计算得到刚体变化参数, 实现碎片较准确拼合。实验结果表明: 本文算法运行时间提高了12%~16%,拼接误差不超过0.750 mm。与传统方法相比, 本文方法能有效地减少因碎片模型受损而造成的拼接缝隙过大、渗透等现象, 拼合误差较小, 高效地实现碎片拼接。
Abstract
The feature extraction of a fracture surface is inaccurate due to the lack of local fragments and damage of the geometric features of ornamentation.Thus, in this paper, a method was proposed to solve this problem based on SURF feature descriptor and Jaccard distance. First, Canny operator was used to extract the contour lines of the debris edges and model surface, and a multi-scale space was constructed to extract the feature points of the fault surface. Second, to solve the high redundancy in constructing feature descriptors and high delay in Euclidean distance calculation, SURF feature descriptors with low redundancy was constructed, and then compared the similarity of feature points with Jaccard distance to determine the optimal adjacency relationship of fragments. Finally, the parameters of rigid body were calculated by ICP method, and the fragments were assembled accurately. The experimental results showed that the running time of the algorithm increased by 12%-16%, and the splicing error was at most 0.750 mm. Compared to the traditional method, the proposed method can effectively reduce the considerably large splicing gap and penetration caused by the damage of the fragment model,as well as the splicing error, and achieve fragment splicing efficiently.
刘晓宁, 狄宏璋, 杨稳, 林芃樾, 王世雄. 基于SURF特征描述符和杰卡德距离的文物碎片拼接[J]. 光学 精密工程, 2020, 28(4): 963. LIU Xiao-ning, DI Hong-zhang, YANG Wen, LIN Peng-yue, WANG Shi-xiong. Mosaic of cultural relics fragments based on SURF feature extraction descriptor and Jaccard distance[J]. Optics and Precision Engineering, 2020, 28(4): 963.