Artificial intelligence-driven visual feature extraction and transfer learning for automatic identification of paintings and photographs

Abstract

The fusion of art using and artificial intelligence (AI) technology has revolutionised the creative landscape, introducing innovative techniques to produce and interpret visual art. AI has emerged as a powerful tool for generating hyper-realistic images and mimicking traditional art styles, raising profound questions about the authenticity and originality of artistic creations. As AI-generated photographs grow increasingly indistinguishable from human-made paintings. The research examines how advanced deep learning techniques enable accurate human vs. AI artwork differentiation through experimental model evaluations. Our research combined the previously trained VGG19 model with a specially developed CNN to discriminate between different image categories. The VGG19 model validated image feature extraction capabilities but the proposed CNN upgraded this performance with domain-based visual art recognition properties. Extensive testing of a curated AI-generated photograph and human-made painting dataset enabled the proposed CNN model to reach a 95% classification success rate, which outperformed the baseline VGG19 model results.