| When people obtain color image information from digital devices,each device has its own color characteristics because of its different color rendering principles.Color transfer between different devices requires a color management system.The most important part of building a color management system is color specialization.Traditional color characterization methods are complex and require a lot of data.Therefore,this paper focuses on the problem of color characterization of digital camera and color printer with artificial neural network method,uses radial basis function（RBF）neural network to study under D65 standard light source,and compares the results with BP neural network,which proves that RBF neural network has lower color difference in the problem of color characterization of digital camera and color printer.The specific work completed is as follows:（1）Research on colorimetric characterization of digital camera based on RBF neural network and BP neural network.Using Canon Power Shot G11 CMOS digital camera to take the X-rite Color Checker Digital SG 140 color card,collecting RGB data of color blocks,then collecting the spectral reflectance of each color block in the color card under the same experimental conditions,and calculating the XYZ value of each color block.Finally,a color specialization model is built using radial basis function neural network and BP neural network.70 sets of data are used as training set and 70 sets of data are used as test set.Under optimal conditions,the final radial basis function neural network has a training set CMC（1:1）（?）chromatic aberration of 1.72and CIE LAB（?）E*abcolor difference is 1.79,test set CMC（1:1）（?）color difference is3.95,CIE LAB（?）E*abcolor difference is 4.89,which is better than BP network.（2）Research on color characteristic of color printer based on RBF and BP neural network.The RGB color space of color printer is divided evenly,and the training and testing data set of 126 color blocks is established,and the XYZ value of each color block is calculated.Finally,RBF and BP neural network are used to establish the model.Finally,under the optimal condition,the training set CMC（1:1）（?）chromatic aberration of RBF network is 1.30,CIE LAB（?）E*ab.The results show that the color difference of（?）E*abis 2.30,the color difference of CMC（1:1）（?）is 3.60,and the color difference of CIE LAB（?）E*abis 3.60.The color difference of（?）E*abis 5.65,which is better than BP neural network.（3）Reverse characterization of digital camera and color printer based on RBF and BP neural network.RBF and BP neural network are used to establish the reverse color characterization model,and Euclidean distance is used as the standard to measure the errors of（?）、（?）and（?）.The training error of digital camera under RBF network is4.21 and the testing error is 8.86,which is better than the error of training and testing using the same digital camera dataset under BP network.And the printer error is too large,finally through the sample distribution analysis of color difference is too large. |
