Measurement of print quality: Joint statistical analysis of paper topography and print defects

Printing papers constitute 45 % of the global paper consumption. As printed products compete with electronic media, printability has become increasingly important as a cost-efficiency factor. It refers to the properties of the paper that ensure trouble-free running through the printing machine and high print quality. The treatment of printability in this thesis is constrained to print quality. The focus is on the dependences between print unevenness caused by small-scale print defects and the local characteristics of paper surface. The dependences are sought using image based measurements. A set of probabilistic analysis methods is proposed and applied to 2D maps of print reflectance and surface topography. A cross-correlation based image registration procedure is first presented to align the images acquired before and after printing. The large amount of data in the aligned images is used to estimate the joint probability density of print reflectance and surface topography. As the probability density deviates from that of multivariate normal distribution, Gaussian mixture modeling is chosen as a flexible parametric representation of the density estimate. The statistical dependence between print reflectance and surface topography is then quantified by mutual information, thus avoiding any assumptions about the linear or nonlinear nature of the dependence. The results from offset printed newsprint and gravure printed supercalendered papers suggest that linear models cannot entirely capture the dependences. The presented analyses are largely concentrated on the low probability tail areas of the distributions that correspond to the abnormally high print reflectance values and deep depressions on the paper surface. Presenting the locations of these extreme values as anomaly maps allows the evaluation of the conditional probability of finding missing ink in regions that exhibit abnormal behavior of surface topography. The results indicate that missing ink in the examined samples is considerably more probable in regions of abnormal surface topography than in randomly selected regions. As expected, however, a majority of the missing ink spots are attributed to other reasons than surface depressions. Anomaly maps are also used to select subsets of the multivariate data, and mutual information is evaluated in these subsets. The dependences expressed by mutual information are weak, but simulations verify that they are statistically significant. The capability of the surface topography values to explain print reflectance is higher in the most abnormal points of topography than overall in the images. The photometric stereo principle applied in this work is a fast method for acquiring surface topography maps that, based on the results, carry information about the printability of the paper. In addition, the probabilistic methods are expected to be applicable to several property maps besides print reflectance and surface topography. The characterization of the probabilistic dependences serves the Bayesian modeling of print quality as a combination of attributes related to the unprinted and printed paper.