Jean-Luc Lesur

Prediction of the spectral reflectance of laser-generated color prints by combination of an optical model and learning methods.

Recent color printing technologies based on the principle of revealing colors on pre-functionalized achromatic supports by laser irradiation offer advanced functionalities, especially for security applications. However, for such technologies, the color prediction is challenging, compared to classic ink-transfer printing systems. The spectral properties of the coloring materials modified by the lasers are not precisely known and may strongly vary, depending on the laser settings, in a nonlinear manner. We show in this study, through the example of the color laser marking (CLM) technology, based on laser bleaching of a mixture of pigments, that the combination of an adapted optical reflectance model and learning methods to get the models parameters enables prediction of the spectral reflectance of any printable color with rather good accuracy. Even though the pigment mixture is formulated from three colored pigments, an analysis of the dimensionality of the spectral space generated by CLM printing, thanks to a principal component analysis decomposition, shows that at least four spectral primaries are needed for accurate spectral reflectance predictions. A polynomial interpolation is then used to relate RGB laser intensities with virtual coordinates of new basis vectors. By studying the influence of the number of calibration patches on the prediction accuracy, we can conclude that a reasonable number of 130 patches are enough to achieve good accuracy in this application.

Visual quality of printed surfaces: Study of homogeneity

This paper introduces a homogeneity assessment method for the printed versions of uniform color images. This parameter has been specifically selected as one of the relevant attributes of printing quality. The method relies on image processing algorithms from a scanned image of the printed surface, especially the computation of gray level co-occurrence matrices and of objective homogeneity attribute inspired of Haralicks parameters. The viewing distance is also taken into account when computing the homogeneity index. Resizing and filtering of the scanned image are performed in order to keep the level of details visible by a standard human observer at short and long distances. The combination of the obtained homogeneity scores on both high and low resolution images provides a homogeneity index, which can be computed for any printed version of a uniform digital image. We tested the method on several hardcopies of a same image, and compared the scores to the empirical evaluations carried out by non-expert observers who were asked to sort the samples and to place them on a metric scale. Our experiments show a good matching between the sorting by the observers and the score computed by our algorithm.