Improving Tone Prediction in Calibration of Electrophotographic Printers by Linear Regression: Using Principal Components to Account for Co-Linearity of Sensor Measurements

Abstract

This work employs principal component regression (PCR) to improve tone prediction accuracy for color electrophotography (EP). During calibration, primary color patches at different half-tone levels are printed on a belt and measured using on-board sensors. Regression models are developed to predict primary color tone values on output media from these on-board sensor measurements. The prediction accuracy of the regression models directly impacts the quality and consistency of color reproduction. Analyses have revealed a high degree of correlation among the on-board sensor measurements of the calibration patches from the same primary color. This indicates that multiple on-board sensor measurements are linearly correlated and using multiple on-board sensor measurements to predict the tone value may improve prediction accuracy if the collinearity of the measurements is taken into consideration. In this study, a PCR-based approach is applied to handle the multicollinear measurements in estimating the regression model coefficients. Experimental results show the proposed PCR models reduce root-mean-squared error by 24.7% over ordinary least-squares regression models.