【作者】 周祎；

【导师】 曹从军；

【作者基本信息】 西安理工大学 ， 制浆造纸工程， 2008， 硕士

【摘要】 颜色空间转换是色彩管理的核心技术,本课题利用EPSON STYLUS PRO9800喷墨打印机打样ECI2002标准随机色靶,选取建模和检验数据,进行CMYK与L~*a~*b~*颜色空间转换模型的研究,为色彩管理颜色空间转换提供更广泛的理论依据以及实践指导。本课题依据四色平面理论,采用最小二乘法建立二元线性回归平面方程,通过相关性、显著性分析,验证了数码打样呈色符合四色平面规律。采用最小二乘法建立网点面积率与对应平面方程系数间的二次曲线方程,并将二次曲线方程代入选定K网点面积率的关于各色平面方程中初步得到选定K网点面积率时的转换方程,对其进行精度检验后确立K网点面积率分别为0%、20%、40%、60%、80%时的CMYK到L~*a~*b~*颜色空间转换方程。采用三次样条插值算法建立选定K网点面积率与对应转换方程中各项系数之间的关系,得到任意K网点面积率时CMYK到L~*a~*b~*颜色空间的正向转换方程,并分析了方程精度。依据BP神经网络理论、GA-BP神经网络理论,按各色块心理明度L~*测量值划分建模数据,确定10段相同结构的BP神经网络并进行训练;考虑到BP神经网络初始权值随机性会影响网络精度及速度等问题,利用遗传算法(GA)优化各段BP神经网络的权值和阈值,作为BP神经网络的初始值进行再训练,经过对比分析,发现经GA优化后的BP神经网络在速度和精度的实现上比单纯的BP神经网络好,同时建立了L~*a~*b~*到CMYK颜色空间的BP神经网络以及GA-BP神经网络反向转换模型。基于数码打样正反转换模型的建立过程为各类印刷方式下色彩管理的颜色空间转换方法提供了更多的参考,具有一定的指导意义。更多还原

【Abstract】 The conversions of color spaces are core technique of color management.The purpose of topic is doing the research about the models of conversion between CMYK and L~*a~*b~* that provides wider theory basis and practical guide on the basis of printing the standard color target ECI2002 obtained from inkjet printer named EPSON STYLUS PRO9800,and choosing the data of building and checking.According to plane theory of four colors,the topic verifies the plane regulation of Digital Proofing paper about four colors by establishment of binary linear regression plane equation built by least-squares method and analyses of equations’ correlation and variance.The quadratic curve equations about relation between dot area and the coefficient of each plane equation are built by least-squares method and then are substituted into the coefficient of plane equations. The conversion equations are initially gained.The conversion equations under given K dot areas with 0%、20%、40%、60%、80%from CMYK to L~*a~*b~* color space can be gotten after check of initial conversion equations’ accuracy.After the relation between K with 0%、20%、40%、60%、80%and each coefficient of conversion equation are found,the positive conversion equation at every K from CMYK to L~*a~*b~* color spaces are built by using cubic spline interpolation algorithm and the accuracy of equations are analyzed.According the theory of BP and GA-BP Neural Network,the data of building model are differentiated by measuring L~* to train ten BP Neural Network with same structure.Because of the random of initial weights in BP Neural Network,the topic uses GA to research best weights and bios of each BP Neural Network in order to apply to BP Neural Network and to retrain BP Neural Network.Through contrast of BP and GA-BP Neural Network,it can be found that GA-BP is better than BP at training speed and accuracy.At the same time,the reverse conversion models of BP and GA-BP Neural Network are built.The process building conversions of color spaces based on Digital Proofing provides more references for the conversion methods between two color spaces of color management under different printing types,which has practical meaning.更多还原