作者: Peter G. Anderson , Henry Kang
DOI: 10.1117/12.57526
关键词:
摘要: In the context of colorimetric matching, intent color scanner and printer calibrations is to characterize devicedependent responses device-independent representations such as CIEXYZ or CIE 1976 L*a*b* (CIELAB). Usually, this accomplished by a two-step process gray balancing matrix transformation, using transfer obtained from multiple polynomial regression. Color calibrations, in particular, are highly nonlinear. Thus, new technique, neural network with Cascade Correlation learning architecture, employed for representing map device values standards. Neural networks are known their capabilities learn nonlinear relationships presented examples. Excellent results obtamed particular net; most training sets, average differences about one Eab. This approach compared approximations ranging 3-term linear fit 14-term cubic equation. The sets indicate that net outperforms approximation. However, comparison not made same ground and generalizations, trained predict relationships it has been with, sometimes rather poor. Nevertheless, very promising tool use in color other technologies general.