【作者】 许宝卉；

【导师】 李言； 刘昕；

【作者基本信息】 西安理工大学 ， 机械工程， 2010， 博士

【摘要】 为了实现开放式印刷系统、电子出版及网络出版系统中精确地色彩传递和再现,国际色彩联盟(International Color Consortium简称ICC)制定了ICC色彩管理标准。彩色显示器是印刷领域印前系统中重要的组成部分,它的呈色性能、色彩空间转换模型精度直接影响彩色图像的显示效果；另一方面,为了将显示屏幕用于印刷系统的软打样,实现真正意义上的“所见即所得”,就要对显示器色彩特性以及色彩空间转换模型进行深入系统地研究。论文在对色彩管理技术的原理、方法、流程、关键技术进行综合分析的基础上,对显示器色彩测试系统进行了设计,提出了从改善采样点空间分布均匀性的角度,来表征输入输出的非线性关系,以提高特征化变换精度的思想,设计了两端细分中间均匀和基于LOG函数变换的测试色靶。通过实验对显示器的色域、通道独立性、空间均匀性、时间均匀性、磷粉恒常性、液晶屏幕的视角、“黑点”对色彩校正的影响以及显示与印刷效果的对比等进行了研究；采用极差理论分析了伽玛值、亮度、色温对输出效果的影响规律,以此为基础提出采用自适应动态伽玛改善图像效果的思想,并给出了相应的调整流程。建立了增加多项式项数、常数项和扩展项的显示器色彩非线性回归转换模型,以解决转换精度问题、误差局部极大或极小问题以及高阶回归出现的精度调节能力下降的问题；并通过改变模型的输出色彩空间分别实现了多项式RGB到XYZ和Lab的转换,转换精度明显提高。在实现显示器三维查找表模型转换的过程中,设计了三棱柱和四面体的插值算法,比较了四种几何体插值算法的转换精度,结果表明：随着立方体不断细分,模型转换精度越来越高,色差都小于3,在人的视觉不敏感范围内。研究了数字驱动值与显示器亮度参数的关系,改变传统阶调为指数曲线的的描述方法,使用二次多项式进行拟合,建立了基于阶调／矩阵的显示器特征化模型；在转换误差较大的情况下,将求解特征矩阵的特征点由最大点转变为中间点,使最终得到的阶调／矩阵模型转换色差大幅度降低,转换精度得以提高。建立了RGB到Lab和Lab到RGB的正反变换BP神经网络模型,探索了反变换过程中色差的估算办法；通过多重训练比较,研究了正反转换网络的最佳层及单元数目,采用结构分解减小了网络规模,提高了训练速度；合理选择基于LOG函数的训练样本、检验样本和均匀训练样本进行比较,最终使BP神经网络实现显示器色彩空间正反转换模型的精度达到了比较高的要求；同时提出了基于神经网络正向变换和查表方案相结合的Lab到RGB的反向变换方法,使正向神经网络模型的应用得到了进一步拓展,实验验证了方法的可行性和有效性。通过以上的研究,使用这些方法实现显示器色彩空间转换,可以满足不同的色彩复制精度要求。其中,三维查找表和神经网络模型所得到的转换色差属于小色差范围,是几种方法中比较理想的方法。这些方法为实现屏幕软打样提供了理论依据,为进一步开发扫描仪、显示器、打印机等全面的色彩管理系统打下了良好的基础、积累了研究经验。更多还原

【Abstract】 The International Color Consortium (ICC) formulated the relative color management standards, in order to meeting accurately communicating and reproducing color information in those systems:opening print system, electronic publishing and network publishing systems. Color display is an important part of the prepress system. And the precision of its colorimetric properties and color space transition model will directly affect the effect of the color images display. Moreover, it can be used for soft-proofing to achieve a real sense of the "What you see is what you get." through display device’s color management. Therefore, the color characteristics and color space transition model of display are studied in this thesis.Based on the comprehensive analysis of color management’s principle, method, process and key technologies, it designed display color test system, proposed a method of characterizing the non-linear relationship between input and output from the view of perfecting distribution uniformity of sampling points’space to improve the characterization transformation accuracy. An ends-subdivided and middle-uniform test color target is designed based on LOG functional transformation.Through the experiment, quantitative analysis variety color characteristics of display, including display color gamut, channel independence, spatial uniformity, time uniformity, phosphorus powder constancy, LCD screen viewing angle and black spots impact on the color correction, as well as comparison of display effect with printing effects, etc. Visual analyzed gamma value, brightness, color temperature impact to the output effect using the range theory. Based on the above research, self-adaption Dynamic Gamma Function is proposed to improve image’s effect, and then a corresponding adjustment process is given. Non-linear regression transformation model is improved effectively using to increase the number of polynomial’s items, the constant term, extensions and other methods, in order to solve the accuracy problem, local maximum or minimum error and regulate ability decline of high-return in accuracy. In addition, the conversion from polynomial of RGB color space to Lab color space was achieved by changing the model’s output color space. Conversion accuracy was compared which of four kinds of geometric interpolation algorithm in the process of model transformation in three-dimension look-up table. The results show that the color difference of model transformation was less than three which was not sensitive to the human visual range.Based on the study of the relationship between digital drive value and the spectral data, tone/matrix model in color management modeling of display characterization is established by using quadratic polynomial fitting. In the case of big conversion error, changing the feature points from the maximum point into the mid-point can make the chromatic difference reduced significantly, so that the conversion accuracy will be improved.BP neural network conversion models from RGB to Lab and Lab to RGB are established, and color difference estimation method is explored in process of reverse-transformation. The best layer numbers of conversion networks are studied through multiple training. Network’s size is decreased by structure decomposing to promote training speed. Training sample, test sample and uniform sample are selected to do comparison so as to realize the requirement of higher accuracy of conversion models. Meanwhile, a method of reverse-transformation from Lab to RGB is proposed based on the forward transformation of neural network and table searching. Experiments verify the reliability and effectiveness of this method.The above research shows that these methods could realize color space transformation of displayer, could satisfy different accuracy requirement of color copy. In which, three-dimension look-up table and neural network conversion models obtained transforming color difference is small, and is a perfect method among kinds of methods relatively. These methods provide theoretical base to realize soft proofing. It is a good foundation and has accumulated practical experience for the further development of such color management systems as scanner, displayer, printer etc.更多还原