PATTERN RECOGNITION APPLIED IN FINE ART AUTHENTICATION

This work investigates pattern recognition tech-niques that can be used to identify the author of fine art paintings based on the content of their digitized images. The methodology proposed consists of four sequential phases: image acquisition, segmentation, feature extrac-tion, and classification. For segmenting brushstrokes it is assumed that the hue is quite uniform in a single brush-stroke. The work uses a wavelet-based technique to measure textures and uses it as a feature to describe the brushstrokes in the painting. Classification is performed through two distinct non-parametric algorithms: Fuzzy KNN and Kohonen Map. A system was built in order to evaluate the proposed approach in the task of discriminat-ing between two different painters. A database consisting of images from 31 paintings was used in the experiments. The system was able to attain a recognition rate greather than 99%