Colour Scanning Calibration – a Comparison of Different Methods

In this master’s project a number of scanner colour calibration methods are evaluated for the purpose of establishing a calibration method for a desktop scanner. A scanner is used when scanning images of homogenous printed areas that subsequently are analysed for the measurement of colour mottle in prints. The master’s project has involved studies of several calibration techniques including artificial neural networks using Bayesian learning. The results show that a method based on artificial neural networks using Bayesian learning, and second a method, based on multiple regression of a 20 term polynomial curve fit, yield similar calibration accuracy. The latter calibration method is recommended because it is simpler to implement and requires less computational effort. A mean accuracy of ∆Eab < 4 was achieved in cases when the evaluated printed samples did not show a high amount of gloss variation, implying that the calibrated scanner-acquired images can be considered to fulfil the requirements needed for the colour mottle application purpose. However, when the samples presented a high amount of gloss variation the accuracy did fall, sometimes dramatically, ∆Eab > 10, suggesting that a scanner in this case is a less suitable input device. Skannerkalibrering – En metodjämförelse