【作者】 付建华；

【导师】 周丰昆；

【作者基本信息】 中国科学院研究生院（长春光学精密机械与物理研究所） ， 光学， 2004， 硕士

【摘要】 随着计算机、媒质、internet的发展，各种彩色图像的精确复现变得越来越重要。许多电子彩色设备如：彩色扫描仪、显示器、打印机的色彩空间都是依赖设备的色彩空间。因而，CIE(Commision Internationale de l’Eclairage)发展起了不依赖于设备的色彩空间来实现颜色在各种颜色设备之间的一致性，所以不依赖设备的颜色复现已经受到人们广泛的注意。 为了实现颜色在依赖于设备的和不依赖于设备的转换，近年来提出来几种色彩空间转换方法如：Neugebauer方程模型法、多元回归法、查表法、神经网络法。 本文主要采用多元回归法和BP神经网络法来实现彩色打印机色彩空间转换。多元回归法主要研究了实验样本数目、阶数对结果的影响，实现了彩色打印机色彩空间转换，避免了使用彩色打印机色彩空间CMYK的复杂性。最后以CIE推荐的色差公式来评价，四阶多元回归取得了较好的预测效果。神经网络主要研究了训练方法、传递函数、隐层数和隐层单元数对结果的影响，最后得出结论三层神经网络即可很好的预测效果。最后比较了两种方法，各有利弊，均可以用于彩色打印机色彩空间的转换。 彩色打印机色彩空间的转换，为计算机公司色彩管理软件的开发奠定了基础。同时它将在医疗、遥感图像、其它颜色设备中都有广泛的应用，是颜色还原的关键环节。更多还原

【Abstract】 With the development of computer, media, internet, the need for accurate and transparent exchange of color content has become critical. Many types of electronic color imaging device color scanner, monitor, printer, all of these devices utilize device-dependent color spaces Therefore, CIE(Commision International del’Eclairage) developed device-independent color spaces to grantee the production of color consistency between various devices. That is, the concept of device-independent color reproduction, has received widespread attention.In order to realize the color space transformation between Device-independent and device-independent, several methods have been proposed for color space transformation. These conclude analytical model method using -Neugebauer equation, the polynomial regression model, look-up tab.le and neural network methods.This article presents study on color space transformation of Color Printer. It presents two methods: polynomial regression and BP neural networks, polynomial regression based on optimization analyzed about the experiment sample number, the polynomial exponent, the selection of test sample and so on., Avoiding the complexity of ink-jet color printer’ s CMYK subtractive color space. This regression method establishes a colorimetric prediction model from input color space transformation. By using the color difference evaluation method recommended by CIE and comparing different exponent experimental result, it is confirmed that the fourth order polynomial had a better prediction effect BP neural networks mainly studied effect on precision training function, transfer function, hidden numbers, hidden neurons and so on . Drew a conclusion three layers BP networks to realized color space transformation of color printers effectively. Finally compared the two methods. There are advantage and disadvantage between polynomial regression and BP neural networks, they are both applied in color space transformation of color printers effectively.It is available for Computer Company soft to develop. This theory will be applied to the fields of Computer Company, medication, the correction in color remote sensing image, andxother color system. It is更多还原