彩色扫描仪的特征化

Characterization of color scanners

阐述了扫描仪特征化的原理和方法。采用多项式回归和人工神经网络技术,建立了扫描仪记录的RGB信息和原影像CIEXYZ及CIELAB色度信息之间的非线性对应关系。对4个常用扫描仪,实验得到了合理的多项式和人工神经网络结构。发现多项式项数为10较为合适,达到的CIELAB平均色差和最大色差分别在1.6～3.0和8～13个色差单位之间;6个隐含层神经元为人工神经网络较为合理的结构,达到的CIELAB平均色差和最大色差分别在1.4～2.6和6～8个色差单位之间。实验结果表明,对常用扫描仪的特征化,可采用多项式回归或人工神经网络技术,其中人工神经网络技术精度较高。

The nonlinear relationship between the scanner RGB signals and original image CIEXYZ or CIELAB values were obtained using the polynomial regression and the artificial neural network procedures. The reasonable structures of the polynomial and the artificial neural network were found for four color scanners. The better number of polynomial terms was 10, reaching the accuracy of 1.6～3.0 CDU (color difference unit) CIELAB average CD (color difference) and 8～13 CDU CIELAB maximal CD. And the reasonable number of hidden cells of the network was 6, reaching the accuracy of 1.4～2.6 CDU CIELAB average CD and 6～8 CDU CIELAB maximal CD. The test results showed that the polynomial regression and the artificial neural network techniques could be used to characterize commonly used color scanners, and the latter has a higher accuracy.