Improving Tone Prediction Accuracy in Calibration for Color Electrophotography Part II – Principal Component Regression

Abstract

This paper presents results of improved tone prediction accuracy in calibration through a principal component analysis based regression approach for color electrophotography (EP). During calibration, multiple color patches of the same primary color at different halftone levels are printed on a belt with black exterior and are measured using on-board sensors. Regression models are often developed to predict the primary color tone values on output media from these on-board sensor measurements. The prediction accuracy of the regression models directly impact the quality and consistency of the calibration. Analyses have revealed a high degree of correlation among the color patch measurements, which results in using multicollinear measurements as explanatory variables during regression analysis to identify model coefficients. It is well known that using collinear explanatory variables during regression analysis will result in suboptimal model coefficients that will degrade the prediction accuracy. In this study, a principle component regression (PCR) approach is applied to tackle the potential issue with collinear measurement data in model coefficient estimation. The experimental results show the resulting PCR models provides 25% improvement on average in root-mean-squared predication accuracy over separate ordinary least square regression models.