Evolutionary automated recognition and characterization of an individual's artistic style

In this paper, we introduce a new image database, consisting of examples of artists' work. Successful classification of this database suggests the capacity to automatically recognize an artist's aesthetic style. We utilize the notion of Transform-based Evolvable Features as a means of evolving features on the space, these features are then evaluated through a standard classifier. We obtain recognition rates for our six artistic styles — relative to images by the other artists and images randomly downloaded from a search engine — of a mean true positive rate of 0.946 and a mean false positive rate of 0.017. Distance metrics designed to indicate the similarity between an arbitrary greyscale image and one of the artistic styles are created from the evolved features. These metrics are capable of ranking control images so that artist-drawn instances appear at the front of the list. We provide evidence that other images ranked as similar by the metric correspond to na¨ıve human notions of similarity as well, suggesting the distance metric could serve as a content-based aesthetic recommender.