Transformation Method Based On Artificial Neural Networks Crt Chroma

The method of CRT color notation conversion, which is very useful in the fields as DTP(desktop publishing), viewing images on the internet and many other color science research works, is one of important topics in color science today. During the past several years , many methods of CRT color calibration have been developed such as 3D-LUT, PLCC, PLVC, LIN-LIN2,LOG-LOG, GOG etc. , and these method have been successfully used in some practical cases. But the all methods before have a common problem of "slow processing speed", and they can not meet the needs of fast speed in video image display. In order to find a more practical method for CRT color conversion, the author has been working on the method of BP-neural-network-based CRT color conversion for many years, and the following results are reached: 1. A method of net-configuration, net-training for BP neural-networks which are used for XYZ to RGB and XYZ to RGB conversion in CRT are developed. 2. A efficient method of color sample choice in training BP neural networks named "LOG method"is put forward and testified by author. 3. The new methods of "grouping training", "derived sample training"and "table-look-up"are proposed and testified , which are practical in the XYZ to RGB and XYZ to RGB conversion. 4. The software and hardware for CRT color conversion based on BP neural networks are realized with a PC1500 spectrometry system and a PC-computer. The experimental results in this paper show the color errors of XYZ to RGB or XYZ to RGB conversion with a set of 1000 training samples and a construction of 3-15-15-15-15-3 of BP neural networks is less than 1 CIELUV unit.