人工神经网络作为打印机色空间转换工具的问题

通过分析大量实测的打印机输入(R,G,B)与输出(L ,a ,b )数据之间的对应关系,发现在一些颜色区域存在着RGB与L a b 之间的二对一甚至多对一的关系。另一方面,一个已知的事实是,当一个神经网络已经建立后,任一个(或一组)输入只能产生一个(或一组)输出。结合以上两个事实得出的结论是,在要求打印机的输出颜色能准确地复现指定颜色的情况下,神经网络不适合作为由指定颜色的L a b 值来寻求打印机的相应输入RGB的工具。

Problems of artificial neural networks used for printer color space conversion

Based on an analysis of the corresponding relationship between the input data (RGB) and the measured output data (L *a *b *) of a color printer, it is found that in some color regions, one set of assigned L *a *b * values may correspond to two or more sets of different input RGB values. On the other side, the fact that an established artificial neural network can only produce one (or one set of) definite output in response to any (or any set of) given input implies that neural networks are not suitable ,at least when high accuracy is expected, to be used in the cases where it is supposed that one input may correspond to more than one outputs. Thus, it is concluded that the neural network is not suitable to be used as a tool to find out a proper set of RGB for a printer to print out an assigned color (L *a *b *), at least this is the case when the accuracy of the said color reproduction is high.