Estimating ink density from colour camera RGB values by the local kernel ridge regression

We present an option for CCD colour camera based ink density measurements in newspaper printing. To solve the task, first, a reflectance spectrum is reconstructed from the CCD colour camera RGB values and then a well-known relation between ink density and the reflectance spectrum of a sample being measured is used to compute the density. To achieve an acceptable spectral reconstruction accuracy, the local kernel ridge regression is employed. The superiority of the local kernel ridge regression over the global regression and the popular ordinary polynomial regression is demonstrated by experimental comparisons. For a database consisting of 250 colour patches printed on newsprint by an ordinary offset printing press, the average spectrum reconstruction error of DE 1⁄4 0:733 and the maximum error DEmax 1⁄4 3:29 was obtained. Such an error proved to be low enough for achieving the average ink density measuring error lower than 0:01D. r 2006 Elsevier Ltd. All rights reserved.