Adaptive polynomial regression for colorimetric scanner calibration using genetic algorithms

Device independent colour representation is a common operation in order to assure colour consistency throughout the colour measurement and reproduction pipeline, i.e., from input to output devices. Different methods exist to convert device dependent colour spaces into device independent colour representation, usually relying on a priori fixed model structures and without imposing limitations on maximum approximation errors. This tends to lead to colour mismatches above the JND in certain regions of the device independent space. Accounting for maximal approximation errors is crucial for some colour application domains, such as in automatic visual inspection and high quality colour reproduction applications. In this paper, a novel colorimetric calibration algorithm for RGB input devices is introduced, using a polynomial modeling strategy. In order to adaptively identify the adequate model structure, i.e. polynomial order and terms, an OCV approach is followed to asymptotically approach the optimal bias-variance trade-off. Furthermore, a weighted least squares approach is applied to, indirectly, limit maximal approximation errors. These weights as well as the optimal model structure are identified using a GA. The performance of the proposed approach is compared with respect to state of the art colorimetric calibration algorithms. The achieved results show that the algorithm exhibits good generalization capabilities and that it is compatible with high quality colour reproduction for all the calibration set sizes tested. Furthermore, better colour space mappings can be obtained by the proposed method compared to the state of the art algorithms.