基于人工神经网络的CCD数码相机颜色特征化

Colorimetric characterization of CCD digital camera based on artificial neural networks

阐述了基于BP神经网络的数码相机特征化方法。采用不同的神经网络结构，建立了数码相机记录的RGB信息和原影像CIEXYZ色度信息之间的非线性对应关系。对NIKOND200数码相机进行了研究，通过实验得到了合理的神经网络结构为3—10—10-3。测试不同的训练样本和测试样本，达到的CIELAB平均色差和最大色差分别为1．9—2．2和6．7—7．4个色差单位。讨论了实验设备的重复性，同时，分析了样本数量对实验结果的影响。实验结果表明：对数码相机的特征化，可采用BP神经网络技术实现较高的精度。

A method is described to characterizing the digital camera. The nonlinear relationship between the RGB signals generated by digital camera and original image CIE XYZ values is obtained using BP neural networks. The reasonable structures（3-10-10-3）of the BP neural networks are found by researching on NIKON 1）200 cameras, yielding a modeling accuracy typically average color difference 1.9 -2.2 AE units and maximal color difference 6.7 ~7.4 AE units. The repeatability of the experimental setup is studied and the influence of the quantity of the samples is also discussed. The experiments resuhs show that the BP neural networks can be used to characterize commonly digital camera.