基于图像配准的古代西夏文活字印刷术鉴别方法

Identification method of movable-type printing of Tangut characters based on image registration

古代西夏活字印本考是现存世界上最早的活字印刷品,然而由于活字聚版不精或活字不规范而造成的字形歪斜、字列不正,对活字印刷术的鉴别研究是一种莫大的考验.针对上述问题,提出一种基于图像配准技术的西夏文活字鉴别方法.该方法首先提取字符图像的边缘特征点,然后根据仿射迭代最近点(iterative closest point,ICP)算法的配准结果来度量字符图像的形状相似性,并在此基础上建立一个连接相似图像的启发树结构,最后通过一致点漂移(coherent point drift,CPD)方法完成图像的非刚体配准.该配准结果能有效地实现西夏文的活字印刷术的鉴别,实验结果表明,与最新方法相比,本文算法具有精度高的特点.

Ancient Tangut printing text is the earliest movable type printing available in the world. However, it is a great challenge for the research of movable type printing identification because its printed characters are crooked and their row is disordered, which is due to its poor and unclear type printing and its non-normativeness. Aimed at above-mentioned problems, a new method is proposed bosed on image registration, which can identify moveable type printing of Tangut texts efficiently. First, the boundary points of each character image are extracted, and the affine iterative closest point （ICP） with bidirectional dis- tance is used to measure the shape similarity of two character point sets. Then, a heuristic tree of Tangut characters is constructed by shape similarity, which connects two point sets with small deformation. Finally, the coherent point drift （CPD） algorithm is adopted to complete non-rigid-registration of the image. This registration result can effectively realize the identification of movable type printing of Tangut text. It is shown by experimental result that compared with the most novel method, the method presented in this article will have a feature of higher accuracy.