Research On The Color Recovery Method Based On RGB-L*a*b* Color Space Transformation

Our daily life and the progress of science are inseparable from the colors. People expect the equipments to reflect more real and richer colors. Recovery method based on color space conversion is one of the key technologies in color restoration, color consistency and color management.In the base of the former researches, this paper does some deep researches on polynomial regression method and BP neural network method. This paper is aimed at using CCD imaging device as a color image collecting equipment to collect data. According to polynomial regression method and BP neural network method, we establish a color correction model based on the input and output information to modify the color of the collected images, so as to realize the accurate color restoration. The research contents are as follows:(1)The RGB to L * a * b * transformation method based on polynomial regression is done further research. The main causes of color distortion are analyzed, and the color restoration scheme is proposed. Then this paper divides the space into sub-domains in accordance with the hue to improve the polynomial regression. Dividing the space into sub-domains in accordance with the hue can reduce the nonlinear of the function, reduce the polynomial terms, and simplify calculations.Pantone contrast experimental results show that the calculated average color difference based on BP neural network is 2.8476, and the difference based on global polynomial regression is 2.857. But after using polynomial regression of each subspace to recover the colors, the average color difference is 2.206, reducing about 23% compared with the above two methods. Recovering the colors by using polynomial regression of each subspace can effectively improve the precision of color restoration.(2)The RGB to L * a * b * transformation method based on BP neural network is done further research. The RGB to L * a * b * transformation methods based on the conventional BP neural network and the genetic algorithm(GA)-BP neural network were analyzed and compared regarding their merits and demerits. A mind evolutionary algorithm(MEA) and adaboost algorithm(Adaboost) are used toimprove the back propagation(BP)-neural network. The captured samples of the color-targets under the color image acquisition equipment are taken as the input data and the standard color data are taken as output. Finally, establish a mapping between the input and output based on mind evolutionary algorithm(MEA)-back propagation(BP)-adaboost algorithm(Adaboost) neural network.In order to verify the extensiveness of the algorithm, the color target IT8.7/2 and color target IT8.7/4 were selected for experiments. The experimental results of the standard color target IT8.7/2 show that the average color difference after training five times based on MEC-BP-Adaboost neural network dropped to 3.49 from 21.108 without correction, and it is about 0.6 to 1.5 smaller than the average color differences of other algorithms. The result of every training was very good, too. The experimental results of the standard color target IT8.7/2 show that the average color difference after training five times based on MEC-BP-Adaboost neural network dropped to 3.49 from 18.919 without correction, and it is about more than 0.2 smaller than the average color differences of other algorithms. The scale of D < 5 E is increased from 0% to 53% and more than 3% higher than other algorithms. The results of 5 times training had smaller difference, and was better than other methods. Experimental results demonstrate that the proposed MEC-BP-Adaboost neural network method achieves much better correction performance with fewer experiments.