【作者】 孙佳石；

【导师】 周丰昆；

【作者基本信息】 中国科学院研究生院（长春光学精密机械与物理研究所） ， 光学， 2005， 博士

【摘要】 敦煌壁画颜色在不同设备(媒质)上复制再现时的颜色校正还原问题,一直是人们关注的热点问题,是文物保护的重要指标之一。敦煌壁画颜色“色彩传递,一致再现”也是敦煌学研究的重要组成部分。 通过分析数码相机到显示器,显示器到彩色打印机的颜色校正还原方法,本文建立了一个全色域范围下的实用可操作的颜色校正还原系统,使壁画图像由数码相机摄取,到计算机处理,再经彩色打印的全系统传输后,彩印出的壁画颜色能够“准确真实”地表现原壁画的颜色。 研究了色域匹配的现有方法,采用了人工神经网络的匹配方法。通过对神经网络的优化设计和学习训练,使该神经网络的权值和阈值得到了充分的信息储存,实现了彩色打印机的L~*a~*b~*色域空间向显示器的RGB色域空间的色域匹配,避免了其它方法的局限性,得到了较好的色域匹配预测值。 由于人工神经网络隐层神经元节点数的确定尚无理论规律可寻,本文提出了应用二次通用旋转组合设计与遗传算法结合,优化得到人工神经网络隐层神经元节点数的方法。本方法也为多隐层人工神经网络隐层神经元节点数的确定奠定了理论基础。 建立了修正三基色值的双隐层人工神经网络,并将修正三基色值的人工神经网络与VC程序嵌套,只需对图像像素三基色值进行修正,便可得到最终的敦煌壁画颜色的校正还原,实现了校正复杂颜色失真过程的简单化。 使用Macbeth ColorChecker Chart对该颜色校正系统进行色差检验,平均色差由校正前的17.1NBS,最大色差37.0NBS,最小色差4.3NBS,降到校正后的平均色差5.3NBS,最大色差12.4NBS,最小色差1.7NBS。取得了较好的客观颜色校正效果,主观评定方面也得到了业内专家的肯定。更多还原

【Abstract】 Dunhuang mural color correction is a hot problem all along when it is reproduced or reappeared on different equipment (medium), it is one of the important indexes on cultural relic protection, achieving the same color reappearance along with the Dunhuang mural transferred, is also an important part of Dunhuang study.By analyzing the color correction method from digital camera to CRT, and from CRT to color printer, the color correction system is established, which can be operated practically under a big color gamut. After transferred through the whole system equipment, which included taking picture with digital camera, computer processing and color copying, the final picture we got from color printer can show the original mural color accurately.Based on the study of color gamut matching method in existence, the matching method of artificial neural network is adopted. With the optimization design and the training of the artificial neural network, ample information is stored in weight and threshold value of the artificial neural network. The color gamut matching is realized from the L*a b color space of the color printer to RGB color space of the CRT, this method can avoid the deficiency of other, and get better forecasting value of color gamut matching.Because of no theory or principle to decide the number of hide-layer neural cell of the artificial neural network, the method combined second general revolving combination design with genetic algorithm is put forward, which can optimize the hide-layer neural cell number of artificial neural network. This method provided theory foundation for the structure establishment of hide-layer neural cell of artificial neural network also.The artificial neural network including two hide-layers is established, which can correct the tristimulus values, the weight and threshold value of this artificial neural network is gained, and the artificial neural network is nested in the computer program compiled by VC language. The color correction of the last Dunhuang mural picture can be finished after the tristimulus values of image pixel are corrected, and the process simplified correcting the complex color distortion has been realized.Using Macbeth ColorChecker Chart, the color difference of this color correction system is tested. Before correction, average color difference is 17.1 NBS, the maximum color difference is 37.0 NBS, and the minimum color difference is 4.3NBS;while after correction, the average color difference is reduced to 5.3 NBS, the maximum color difference is 12.4 NBS, and the minimum color difference is 1.7 NBS. Ideal objective effect of color correction has been gotten and expert in this domain has praised the work.更多还原