节点文献
基于人工神经网络的彩色扫描仪特征化
Characterization of Colorscanners Based on Artificial Neural Network
【作者】 李斌；
【导师】 张逸新；
【作者基本信息】 江南大学 ， 印刷工程与媒体技术， 2012， 硕士
【摘要】 本文通过对彩色扫描仪特征化各种方法的调研和支持向量回归机(SVR)用于非线性回归问题数值计算方法的特点分析,研究分析了采用SVR方法实现彩色扫描仪RGB色空间到与设备无关的CIELAB色空间非线性转换的可行性,并建立了扫描仪从RGB色空间到CIELAB色空间转换的特征化数值实验模型。为了验证特征化模型的色空间转换精度,本文采用MATLABR2009进行数值模拟实验,数值实验中选用IT8.7/2标准色卡作为色靶,选用CIELAB和CIEDE2000色差公式作为色差评价标准进行色差计算。实验中先用扫描仪和色度仪测得IT8.7/2色卡上各色块的RGB值和CIELAB值,然后将288个色块分为2组,分别进行训练和测试,最后采用CIELAB、CIEDE2000色差公式对测试组实验结果进行色差计算。研究表明: L?、a?、b?的预测值与实际值之间的回归相关性达到99%以上。由CIELAB色差公式计算得到的色差平均值、最大值和最小值分别为2.3143、5.7917和0.5073;由CIEDE2000色差公式计算得到的色差平均值、最大值和最小值分别为2.6597、5.2194和0.2065。实验结果表明,SVR可以很好的实现RGB色空间到CIELAB色空间的非线性转换,模型精度符合印刷色彩管理要求,体现了SVR在非线性回归问题中精度和时间的优势,可用于扫描仪特征化管理中。
【Abstract】 We research the principle of color scanner characterization and Support Vector Regression (SVR) respectively and then analyze the possibility of nonlinear transformation from scanner RGB color space to CIELAB color space based on Support Vector Regression (SVR). We built a model of scanner RGB color space to CIELAB color space based on SVR.To prove the accuracy of the model, choose IT8.7/2 color card as color target and make data simulation with MATLABR2009a software and figure out the color differences using CIELAB color difference formula and CIEDE2000 color difference formula. In the Experiment ,we use the scanner and chromaticity instrument to measure the RGB values and CIE values of all color lumps and divide the 288 color lumps into two groups (training and testing group) .The experimental results show that the similarity is more than 99% between predicted values and true values of L?、a?、b? . Based on CIELAB color difference formula, the average、the maximum and the minimum color differences of test set are 2.3143、5.7917 and 0.5073. Based on CIEDE2000 color difference formula, the average、the maximum and the minimum color differences of test set are 2.6597、5.2194 and 0.2065. From the results, we can make a conclusion that SVR can realize the nonlinear transformation from scanner RGB color space to CIELAB color space and the model satisfies the accuracy of scanner characterization. Therefore, SVR is a very promising tool for using in color scanner characterization management.
【Key words】 scanner； characterization of scanner； artificial neural network； support vector machine； support vector regression； CIEDE2000；