WATERCOLOUR IDENTIFICATION BASED ON MACHINE VISION ANALYSIS

In this paper we present a novel image-based approach forcharacterizing and classifying blue and green pigments asused in paintings based on their optical properties. Our aimis to develop a non-destructive method that can characterizeand classify watercolours through a statisticalapproach which recognizes both the probabilistic nature ofthe optical watercolour information and the form in whichwe should express the results. Furthermore, this study alsocombines historical information and the statistical computationof the watercolour optical behaviour.Our method is non-invasive, does not involve sampling andcan be applied in situ. It is based on the optical propertiesof pigments as well as the correlation of pigment variations.Such variations are studied using co-occurrencematrices that capture the behaviour of the pigments duringthe painting process. Furthermore, the optical response ofwatercolour pigments can be represented using a mixtureof Gaussian functions, and can be classified using aBayesian decision rule.