【作者】 孙静；

【导师】 曹从军；

【作者基本信息】 西安理工大学 ， 制浆造纸工程， 2009， 硕士

【摘要】 现代先进的ICC色彩管理技术,是以色彩空间变化为核心的新型色彩控制方法,其中CMYK和L*a*b*颜色空间的转换在色彩管理中具有广阔的理论研究价值和实用价值。课题基于数码打样实验,输出在Photoshop中设计的色靶,选取建模数据和检验数据,基于平面理论建立CMYK到L*a*b*颜色空间的正向转换模型；分别基于GA—BP神经网络、广义回归神经网络(GRNN)建立L*a*b*到CMYK颜色空间反向转换模型。依据四色网点的平面呈色规律,利用数理统计中的二元线性回归的方法建立K网点面积率以及CMY某一彩色网点面积率一定,其他两彩色网点面积率变化的平面方程,再利用最小二乘抛物线法得到平面方程每个系数与网点面积率的二次拟合曲线,最终建立起K网点面积率分别为0%,10%,20%,30%,40%,50%,60%,70%,80%,85%,90%,95%,98%的颜色空间转换方程。经过色差分析,发现特定K值下颜色空间转换方程具有较高精度,且K=50%以后的转换方程比K=50%以前的方程转换精度高,这可能是因为亮调区域受测量值精度的影响。然后通过三次样条插值算法,建立起任意的K网点面积率下的颜色空间转换方程,方程也可达到较高精度,且其精度受特定K值下颜色空间转换方程精度的影响。因此增加网点面积率K=50%以后的数据,基于平面理论可以实现任意K网点面积率下CMYK到L*a*b*颜色空间更精确的转换。采用GA—BP神经网络建立颜色空间反向转换模型时,隐层和隐层神经元数的确定和调整过程很繁琐,且模型的预测精度不高,因此GA—BP神经网络应用于L*a*b*到CMYK的转换还是存在不少问题的。而基于GRNN进行的颜色模型变换研究可知该网络在学习样本确定后,则相应的网络结构和各神经元之间的连接权值也随之确定,网络的训练实际上只需确定SPREAD值,网络训练速度快、容易实现且模型的精度较高。更多还原

【Abstract】 The conversions of color spaces are core technique of modern ICC color management and the study of color space conversion algorithm between L*a*b* and CMYK is value both in theory and application. The building and testing data are obtained by outputting target designed in Photoshop based on digital proofing. The forward color space conversion model between CMYK and L*a*b* is based on plane theory and the reverse models are based on BP Neural Network and Generalized Regression Neural Network respectively.According to plane theory of four colors the plane equation is build by binary linear regression method. Then the quadratic curve equations about relation between the coefficient of each plane equation and dot area are built by least-squares method. The conversion equations under given K dot areas with 0%,10%,20%,30%,40%,50%,60%,70%,80%,85%, 90%,95%,98% are gained. After computation it can be found that the conversion equations under given K dot areas have high accuracy and the equations after K=50% have higher accuracy than those before K=50% which is probably because the light area is affected by measure data. Finally the conversion equations at every K from CMYK to L*a*b* color spaces are built by using cubic spline interpolation algorithm and they also have high accuracy which is affected by accuracy of equations under given K.The number of hidden layers and neurons of hidden layers are difficult to decide when reverse model is built based on GA-BP Neural Network. And the model has poor accuracy so there are many problems of building reverse model based on GA-BP Neural Network. The study of reverse model based on GRNN shows that if the building data is determined the corresponding network is also determined and the training of network is the process to determine the SPREAD value. GRNN grains an advantage over GA-BP network weather in training conveniency, speed or precision.更多还原