基于积分几何的文物表面规律性显著特征提取

Regular Salient Features Extraction on Surface of Cultural Relics Based on Integral Geometry

有浮雕或印记的文物表面上存在一些规律的或重复的显著特征,提取这样的特征结构可用于碎片的识别、拼接和修复中,因此提出了一种基于积分不变量和几何模板的特征提取方法,利用先验知识以及积分不变量对高噪声鲁棒、多尺度的特性来提取特征。首先计算碎片表面每个顶点的积分不变量值,其次用改进的K-均值算法和扩张搜索方法得到局部几何特征,最后结合考古学家的经验、历史知识以及文物出处等先验知识来构建几何模板。将复杂的、部分缺损并带有噪声的秦始皇兵马俑碎片作为实验数据,成功地提取到铠甲库的30个模板。将模板应用于铠甲碎片的识别中,对346个标定的碎片识别,识别率达到86.42%。实验结果表明了此方法的有效性和鲁棒性。

There are some regular or repeated salient features on the surface of cultural relics which is embossed or marked and such structures can be used in the work of pieces＇ identification, matching and repairing, so a feature extraction method based on integral invariants and geometric templates was proposed. This method utilized prior knowledge and integral invariants＇ robustness of high-noise and multi-scale to extract the features. Firstly, values of integral invariants were calculated at all vertexes on a piece＇s surface. Then local shape features were obtained using an improved K-means algorithm and the expanded search method. At last, geometric templates were constructed with prior knowledge such as archaeologists＇ experience, historical knowledge, the provenance of the cultural relics and so on. Taking the complex and partial lacked Terra-cotta Warrior＇s pieces with noise as experimental data, 30 templates in armor library were extracted successfully. Applying them in the experiment of armor pieces＇ identification, which includes 346 labeled pieces, the recognition rate is 86.42%. The results show that the method is effective and robust.