基于KL展开的古铜镜图象相似性评估

Similarity Evaluation of Bronze Mirror Images Based on KL Expansion

古铜镜在人文考古学上有着极为重要的研究价值，而古铜镜图象间的相似性评估则是解决古铜镜资料人工整理分析中所存在的诸多缺陷、实现其大规模定量解析的关键问题之一。此文探讨了ＫＬ展开在古铜镜图象相似性评估中的可适用性，明确了古铜镜图象的规范化处理以及基于ＫＬ展开的特征镜空间构成过程，提出了３种基于“类别”概念的相似性评估标准，并根据实验结果对各种相似性评估标准进行了详细的比较分析。实验结果表明，文中所提出的３种基于“类别”概念的相似性评估标准比传统的距离评估标准更为有效，在较高维数的ＫＬ特征镜空间上评估正确率可达到８５％以上。

Bronze mirrors have very important research values on archaeology, and the similarity evaluation of bronze mirror images is the key to solving the problems in manual classification and realizing the largescale quantitative analysis of bronze mirror images. This paper discussed the feasibility of KL expansion for similarity evaluation of mirror images, specified the normalization of mirror images and the procedure of eigenspace construction through KL expansion, and proposed 3 similarity evaluation standards based on the 'class' conception. Experimental comparisons and analyses of these evaluation standards show that, the 3 proposed evaluation standards based on the 'class' conception are more effective than the conventional distance standard, and their evaluation accuracy reached higher than 85% under eigenspace of higher dimension.