该研究目的为提出一种面向我国各类文物体系的典型色彩光谱样本集构建方法。 以具有1700多年历史的世界级佛教艺术文化遗产敦煌莫高窟壁画彩绘为典型研究对象, 以保证文物色彩传递过程中的色彩恒常性为主要创新, 最终从视觉感知角度提出基于色彩恒常性的敦煌典型色彩光谱样本集构建方法。 在具体实施过程中, 在敦煌研究院美术所及保护所工作人员的配合与指导下, 结合现有敦煌学研究成果, 研究选取了48类敦煌典型颜料并设计制备了240个代表性色彩样本。 利用分光光度计获取其在360～750 nm范围内反射光谱信息, 通过聚类分析法对原始色彩光谱样本集进行分类, 以6类典型光源场景条件下色彩易变性指数（color inconstancy index）的综合量值为依据, 结合Wilcoxon符号秩检验, 确定分类数量及各子类的代表性样本, 初步构建光谱样本集。 随后, 从色域完整性角度对初步建立的样本集进行补充, 最终构建由100个样本组成的全色域光谱数据集。 分析结果显示, 该样本集在99%置信条件下的色彩易变性指数可低至3.28, 且均匀分布于敦煌色彩体系的整个色域, 因而可以为敦煌色彩再现过程中合理的选用具有高恒常性水平的色彩提供数据参考及方法依据。

The present paper aims at developping a method to reasonably set up the typical spectral color dataset for different kinds of Chinese cultural heritage in color rendering process. The world famous wall paintings dating from more than 1700 years ago in Dunhuang Mogao Grottoes was taken as typical case in this research. In order to maintain the color constancy during the color rendering workflow of Dunhuang culture relics, a chromatic adaptation based method for developping the spectral dataset of typical colors for those wall paintings was proposed from the view point of human vision perception ability. Under the help and guidance of researchers in the art-research institution and protection-research institution of Dunhuang Academy and according to the existing research achievement of Dunhuang Research in the past years, 48 typical known Dunhuang pigments were chosen and 240 representative color samples were made with reflective spectral ranging from 360 to 750 nm was acquired by a spectrometer. In order to find the typical colors of the above mentioned color samples, the original dataset was devided into several subgroups by clustering analysis. The grouping number, together with the most typical samples for each subgroup which made up the firstly built typical color dataset, was determined by wilcoxon signed rank test according to the color inconstancy index comprehensively calculated under 6 typical illuminating conditions. Considering the completeness of gamut of Dunhuang wall paintings, 8 complementary colors was determined and finally the typical spectral color dataset was built up which contains 100 representative spectral colors. The analytical calculating results show that the median color inconstancy index of the built dataset in 99% confidence level by wilcoxon signed rank test was 3.28 and the 100 colors are distributing in the whole gamut uniformly, which ensures that this dataset can provide reasonable reference for choosing the color with highest color constancy during the color rendering process of Dunhuang cultural heritage.