基于拓扑与网格双特征的铭文图形识别方法  

Recognition Method of Inscription Graphics Based on Dual Features of Topology and Mesh

作　　者：刘文腾 王慧琴[1] 王可[1] 王展 Liu Wenteng;Wang Huiqin;Wang Ke;Wang Zhan(Collge of Information and Control Engineering,Xi’an University of Architecture and Technology,Xi’an,Shaanxi 710055,China;Shaanxi Institute for the Preservation of Cultural Heritage,Xi’an,Shaanxi 710075,China)

机构地区：[1]西安建筑科技大学信息与控制工程学院,陕西西安710055 [2]陕西省文物保护研究院,陕西西安710075

出　　处：《激光与光电子学进展》2022年第4期221-231,共11页Laser & Optoelectronics Progress

基　　金：教育部归国留学人员科研扶持项目(K05055);陕西省自然科学基金(2021JM-377);陕西省科技厅国际科技合作计划(2020KW-012);陕西省教育厅重点项目高端智库(18JT006);西安市科技局项目(GXYD10.1)。

摘　　要：青铜器铭文图像有效特征的提取是进行铭文识别的关键步骤,针对以图像为信息载体的铭文特征提取方法由于特征维度高、特征向量复杂而识别准确度低的问题,提出了一种基于拓扑与网格双特征的铭文图形集成学习识别方法。以图形为铭文特征的表征,所提方法提取拓扑特征和7维文字结构图形特征,有效描述了铭文文字的结构信息。在此基础上,所提方法利用降维后铭文全局结构信息和局部结构信息的8维4方向弹性网格特征,解决了提取铭文图像特征导致的特征向量维度高的问题。最后,以拓扑特征和弹性网格特征作为集成学习样本的特征向量,利用Bagging方法对特征向量敏感程度不同的机器学习分类器进行集成,提升模型训练效率、提高识别精度。实验结果表明,与图像特征提取方法相比,所提方法对铭文识别准确率提高了15.54个百分点,并且铭文特征向量维度及运行时间大幅度降低。Extracting the effective features of bronze inscription image is the key step of inscription recognition.Aiming at the problem of low recognition accuracy of inscription feature extraction method with image as information carrier due to high feature dimension and complex feature vector, an integrated learning and recognition method of inscription graphics based on the dual features of topology and mesh is proposed. Taking graphics as the representation of inscriptions, the proposed method extracts topological features and 7-dimensional text structure graphic features, which effectively describes the structure information of inscription text. On this basis, the proposed method uses the 8-dimensional and 4-direction elastic mesh features of the global and local structure information of the inscription after dimensionality reduction to solve the problem of high dimension of the feature vector caused by the extraction of the image features of the inscription. Finally, taking topological features and elastic mesh features as the feature vectors of integrated learning samples, Bagging method is used to integrate machine learning classifiers with different sensitivity of feature vectors, so as to improve the model training efficiency and recognition accuracy.The experimental results show that compared with the image feature extraction method, the proposed method improves the accuracy of inscription recognition by 15. 54 percent,and the dimension of inscription feature vector and running time are greatly reduced.

关 键 词：图像处理 青铜器铭文 拓扑特征 网格特征 机器学习 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]