基于人工智能的甲骨文识别技术与字形数据库构建

AI-driven Oracle Bone Inscription Recognition and Glyph Database Development

已有的甲骨文工具书在整理模式、编排方式、检索方法、呈现手段等方面积累了经验,但也有字形分合无标准、检索困难、隶定方式混乱等现象。纸制工具书不利于复用、优化和改进,且无法对每个字形的相关资料穷尽性枚举并提供相关的上下文信息。利用新一代人工智能技术处理数量庞大、表义复杂、字形多变的甲骨文字,训练深度学习模型,实现在全张拓片上对甲骨文字逐一自动定位、识别,可依据字形自动进行分类,在此基础上建设材料完整、语段信息方便查看、开放式编辑的甲骨文字形数据库。进一步设计相关甲骨文识别与相似性检索软件,使字形数据库成为以图片检索甲骨文的便利工具。

Researchers have achieved significant progress in collecting and investigating the oracle bone inscriptions(OBIs),and the research outcomes were usually published and spread in form of reference books.The community has accumulated lots of experiences in the arrangement,retrieval and representation of OBIs.Yet,there still exists many other problems,such as the inconsistency in the separation and union of the character patterns,the difficulty in retrieving OBIs,and the discrepancy in’liDing’(vectoring-drawing)the OBI characters/glyphs.Reference books are easy to be disseminated,but the results summarized in the books are hard to be reused and improved;moreover,due to the page size limit,these books cannot include the complete material and related contextual information of the OBIs.To this end,in the work we adopt the new generation of AI techniques for OBI recognition and database development.In specific,we will train deep learning models for text detection and recognition over the OBI rubbings.Upon this,we will design algorithms to automatically build the glyph database with all the materials and context information,using unsupervised or semi-supervised image similarity computation methodology.Finally,we will also develop a software to facilitate the recognition and retrieval of the OBI characters and the related materials.