This article presents a new methodology for the color accuracy optimization of a two-dimensional digital reproduction. Selecting a training set referring to the colorimetric content of the object to be reproduced results in a significant improvement of the color accuracy of an RGB reproduction. Some authors have developed methodologies for color accuracy optimization that involve the creation of specific custom-made reference targets. The methodology presented here does not involve the creation of custom-made targets, as the reference colors are selected directly on the object. In fact, a clusterization is performed on the RGB image of the object and a set of representative colors is achieved. For each RGB representative color, the corresponding CIELAB value is measured and a training set is obtained that can be used to define a transformation that maps all RGB values into CIELAB values. The experiments conducted show that using this methodology considerably improves color accuracy.