【作者】 盖丽英；

【导师】 成刚虎；

【作者基本信息】 西安理工大学 ， 制浆造纸工程， 2008， 硕士

【摘要】 扫描仪是印刷系统中最主要的一种图像获取设备,由于受到设备颜色特性、观察条件、色域范围等诸多因素的影响,一幅彩色图像原稿在扫描过程中,阶调层次和色彩信息常常会丢失,不能够忠实再现原稿,直接影响着后续工作的正常进行。因此,对扫描仪进行以ICC为标准的色彩特性化研究是十分必要的。选取AGFA DUOSCAN T2500为主要研究对象,分别对标准IT8.7/2色标和自设色靶进行扫描实验和数据处理,采用多元回归方法和BP神经网络改进方法分别建立了扫描仪的ICC特性文件,并分析了目标色空间、多元回归项及项数、不同的变换方式、建模样本数、隐层神经元数目、训练时间、网络层数等参数对模型精度的影响。自行开发了创建扫描仪ICC特性文件的软件,并进行了软件的色彩管理效果的评价。研究结果表明:将色度信号线性化引入到多元回归算法中,能够提高模型的精度,20项的M5模型的精度最高;增加常数项可以提高以CIE LAB为输出空间的模型精度,增加交叉项可以提高以CIE XYZ为输出空间的模型精度;隐层神经元数目在25-35之间均可以达到较好的精度,而采用三层3-30-3的一步正割的BP训练方法可以较好得改进BP神经网络训练精度。合二为一的基于混合模型的扫描仪特性文件综合了基于N维查表模型的输入设备特性文件和基于三维矩阵模型的输入设备特性文件两者的优点,增强了数据转换的确定性,提高了特性文件的精度。以CIE LAB为PCS空间的基于混合模型的特性文件色彩管理效果最好,优于ProfileMaker的精度。更多还原

【Abstract】 Scanner is one of the most important devices for obtaining images in printing system. Gradation of tone and color information of a color picture original often disappear in the scanning process and original manuscript can’t be re-created trustily in virtue of devices’ colorimetric characteristic, observation conditions, colorimetric ration and other influencing factors. As a result, the follow-up process can’t be preceded regularly. Therefore, it is very necessary to study scanner’s colorimetric characterization in accordance with standard ICC.AGFA DUOScan T2500 is the principal object of study. Standard IT8.7/2 and designed color target were scanned and handled dates individually. ICC profiles of scanner were established individually in the method of polynomial regression algorithm and BP neural network theory. Parameters influencing model precision were analyzed. These parameters include target color space, polynomial regression items, items number, different conversion types, modeling specimen quantity, number of concealed gradation, training time, number of network gradation and so on. A soft-ware establishing a scanner ICC profiles was exploited voluntarily and the color management outcome of the exploited software was estimated.It was indicated: Model precision was enhanced by injecting chromatic signal linear into polynomial regression algorithm and the precision of 20th model M5 is the best. Model precision of output space CIE LAB can be enhanced by adding constant terms and model precision of output space CIE XYZ can be enhanced by adding intersects. The number of concealed gradation neuron between 25 to30 can obtain better precision and training precision of BP neural network can be enhanced more validly by using three gradations 3-30-3 one step secant BP training way. The scanner profiles based on mixed models integrate N-component LUT-based input profiles advantage and Three-component matrix-based input profiles advantage. The decision of date transformation and the precision of profiles were enhanced. The profiles based on mixed models and output space CIE LAB PCS whose color management outcome is the best and whose precision is better than Profile Maker.更多还原