Characterization Of Color Scanners Based On The Nonlinear Polynomialreg Ression Model

As the progress of science and technology in life, color scanners have become an essential part in the field of color duplication, which directly affect the accuracy of the color reappearance quality. However, due to the different mechanism in color scanners, the same pictures differ markedly each other toward scanners. Therefore, in order to materialize the quondam color images truly, we must use chromaticity characteristic method to standard identification function of color scanners.This paper puts forward of multivariate nonlinear regression model based on the color of the optimization of the characterization scanner thought, combining with the experimental results in least squares calibration further, we get the eventual results that regression model using RGB to CIE-LAB color space has slightly higher accuracy than RGB output to the CIE-XYZ color space. From CIE-LAB as color output space, third-order and fourth-order in training samples polynomial average chromatic error is:1.1920,0.8382; the test samples of third-order and fourth-order polynomial average chromatic error is:1.3306,1.7810. With the increasing of the order numbers, the training samples in the higher order numbers have higher precision under nonlinear regression optimization, meanwhile the test samples’precision drop and polynomial equation presents sick, generalization ability declines.Compared with the composition and distribution of polynomial coefficients, the polynomial regression still needs more accurate matching. For the independent variable exists between the high linear relationship, we use partial least squares regression model to further optimization. Using four-order polynomial optimization model, Training samples’and the test samples’average chromatic error is:0.7341,1.2110. Contrast to the least squares regression model, the partial least squares not only directly reduce the error of the sample selection and test sample chromatic error is also reduced. In other words, partial least squares solve the state of sick for too much composition in a polynomial equation; therefore enhance generalization ability.