Patch-Based Oil Painting Forgery Detection Based on Brushstroke Analysis Using Generative Adversarial Networks and Depth Visualization

Art authentication has traditionally required deep expertise and knowledge of an artist’s work. Recently, computer vision algorithms have shown promise in image processing tasks; however, creating an automated model for painting authentication remains a challenge in art preservation and history. The challenge is heightened as forgers cleverly create artworks that imitate the original artist’s unique brushstroke signature while introducing new content. To address this and to emphasize the importance of the artist’s unique brushstroke signature, we present a model leveraging conditional Generative Adversarial Networks, trained on in-depth visualization of an artist’s brushstroke style. Two methods including forgery scores computation using frequency analysis and trained discriminator models are proposed to identify counterfeiting. To visualize the depth of the brushstrokes datasets, we use the Reflectance Transformation Imaging technique. We evaluate the authentication of oil paintings by a contemporary Korean artist at the level of image patches in two scenarios. First, we distinguish an original painting from its corresponding counterfeit. Second, we detect a forged painting with creative content that mimics the original brushstroke signature. Results suggest that Generative Adversarial Networks trained on in-depth information have the potential to augment traditional methods in art authentication when utilized by connoisseurs.