Study on color model conversion for camera with neural network based on the combination between second general revolving combination design and genetic algorithm

摘要

Munsell color system is selected to establish the mutual conversion between RGB and L~*a~*b~* color model for camera. The color luminance meter and CCD camera synchronously measure the same color card, the color picture captured from CCD camera is expressed for RGB value as the input of neural network; XYZ value is gotten from the color luminance meter, and the L~*a~*b~* value converted from XYZ value is regarded as the real color value of target card, namely the output of neural network. The neural network of two hidden-layers is considered, so the second general revolving combination design is introduced into optimizing the structure of neural network, which can carry optimization through unifying project design, data processing and the precision of regression equation. Their mathematics model of encoding space is gained, and the significance inspection shows the confidence degree of regression equation is 99%. The mathematics model is optimized by genetic algorithm, optimization solution is gotten, and function value of the goal is 0.0007168. The neural network of the optimization solution is trained; the training error is 0.000748566, which the difference is not obvious comparing with forecast resu it can show that the method combining second general revolving combination design with genetic algorithm can optimize the hidden-layer structure of neural network. Using the data of testing set to test this network and calculating the color difference between forecast value and true value, the maximum is 5.6357 NBS, the minimum is 0.5311 NBS, and the average of color difference is 3.1744NBS.