【作者】 柴冰华；

【导师】 赵达尊； 廖宁放；

【作者基本信息】 北京理工大学 ， 光学工程， 2006， 博士

【摘要】 1931年制定的CIE(国际照明委员会)色度系统,为颜色科学的理论及其应用技术的发展奠定了基础。然而,进入二十世纪80年代,以计算机技术为代表的信息技术得到了迅猛的发展,传统的色度学理论已经难以解决各种颜色信息交流系统中颜色信息的保真传递和交流问题,由此在颜色科学领域引发了一场持续的色貌模型研究热潮。为了实现跨媒体颜色保真复制,色貌模型是必不可少的一个环节。CIE TC1- 34在1997年建立了色貌模型的统一简化版本—CIECAM97s,经过4年多的测试检验,CIE又于2002年公布了CIECAM97s修正版本CIECAM02,补充了一些信息并更加趋向于实用。但是我们注意到,在国际产业界,迄今为止尚未有任何一种色貌模型被相关领域的颜色管理系统所采用。例如,致力于解决跨媒体之间彩色信息保真交流问题的国际色彩联盟ICC(International Color Consortium)从未把任何一种色貌模型应用在ICC的色彩描述头文件(profile)中。因此可认为,传统色貌模型的理论及其应用仍处于发展、试验阶段。基于传统色貌模型本身的复杂性及其在应用中的局限性,我们提出了“关联色貌模型”这一概念,直接将媒体的颜色驱动参数与色貌属性参数(包括照明、背景、环境因素等)联系起来。“关联色貌模型”既可以直接应用于各种自成体系的闭环颜色控制系统中,又可以与目前国际上流行的开放式系统中的“色彩描述文件”融合为一体,从而使整个系统得以简化。从色貌本身的特性和实际应用时的运行速度考虑,采用神经网络方法来实现这个模型。基于传统色貌模型在颜色复制应用时的复杂性,本文提出了一种“视觉匹配”的方法,来获得训练神经网络所需要的样本数据对,并用CIECAM02模型去检验“视觉匹配”方法所获得的实验数据的可靠性(同时也是对CIECAM02的检验),然后通过神经网络方法进行泛化,实现基于神经网络的“关联色貌模型”模块。为了提高神经网络泛化的精度,我们对训练样本进行了优化排序、色空间转换等处理,并提出和实现了各种主观评价与仪器测量相结合的评价方法。同时,利用这种“视觉匹配”和人工神经网络相结合的方法来实现CRT特性化。这种方法由于包含了人眼的色适应因素,所以是一种给定观察条件下基于色貌的视觉特性化方法。在此基础上,建构了一条简洁的颜色复制数据链,实现了基于该方法从硬拷贝NCS色卡到CRT的复制。另外,鉴于CIECAM02模型只能对处于无彩色背景下的颜色实现准确的色貌预测,还提出了一种对彩色背景下的输入颜色进行相对简单的预处理方法,以便当把预处理得到的数据作为输入时,CIECAM02也可以准确预测。基于对立色理论,通过心理物理视觉实验,获得了一个预处理的色诱导公式。通过有预处理和无预处理的结果比较看出,预处理可以改善预测精度,说明预处理概念提出应该是有意义和有价值的。更多还原

【Abstract】 The 1931 CIE (COMMISSION INTERNATIONALE DE L′èCLAIRAGE) colorimetric system has established a basis for the development of theory and application technology of color science. However, during the eighties of the 20th century, while the information technology represented by computers has obtained rapid progress, the problems of true color information transformation and communication in various color information communication systems could not be resolved by the traditional colorimetric theory, thereout a durative research upsurge of color appearance model (CAM) has been brought in the field of color science.For realizing the high perceptional quality cross-media color reproduction, CAMs are necessary. CIE TC1-34 set up a simplized edition of color appearance model― CIECAM97s in 1997. After lots of tests for four years, CIE published the CIECAM02 in 2002, which is a revised edition of CIECAM97s by adding some supplements and making the latter more practical. But as we have noticed, in the color related industries, no CAM has been adopted by any color management system so far. For instance, ICC (International Color Consortium), which is devoting to cross-media color information communication, has never adopted any CAM in ICC profiles. Therefore, it is considered that the theory and application of traditional CAMs are still at the stage of development and testing.Considering the application complexity of traditional CAMs, we proposed a conception of“medium-dependent CAM”, which connects directly the driving parameters of color media with color appearance attributes (including illumination, background, surrounds etc.). The“medium- dependent CAM”can be applied to various close loop control systems of, as well as integrated with the prevalent“ICC profile”of open systems. Accordingly, the whole system gets simplized. Based on the conditions of characteristics of CAMs and the operation speed in practical application, neural networks were adopted to realize this model.Due to the complexity of the traditional CAM in color reproduction applications, we proposed a“visual matching”method to get sample data pairs needed by the training of neural networks, and tested the reliability of experimental data from“visual matching”using CIECAM02. Then these data were generalized by neural networks to establish a neural network based module of“medium-dependent CAM”. To improve the generalization capability of neural networks, the arrangement order of training samples were optimized and the color spaces transform was implemented. We put forward and realized some combination assessment methods of various subjective assessments and instrument measurements. At the same time, we also realized CRT characterization using“visual matching”and neural networks. Since this kind of method involves the chromatic adaptation factor of human eyes, it is a color appearance based characterization method under some given viewing conditions. A succinct color reproduction data chain has been constructed, and the reproductions from NCS color chips (hardcopy) to CRT displays (softcopy) have been realized according to this method.In addition, due to the fact that CIECAM02 can only make accurate color appearance prediction for colors under achromatic backgrounds, has have proposed a comparatively simple pre-process method for the input color under chromatic background so that CIECAM02 can make accurate prediction when the data obtained by the pre-processing are taken as inputs. Based on opponent-colors theory, we obtained a color induction formula for pre-processing through psychophysical visual experiments. By comparing the results from with pre-processing and without pre-processing, it is clear that the pre-processing can improve the prediction accuracy, and this means that the proposition of pre-processing should be meaningful and valuable.更多还原