A Fine-Grained Artist Identification Method for Authentication and Attribution of Drawings Using Hatching Lines

Abstract

Fine-grained image recognition (FGIR) addresses the challenging task of distinguishing between visually similar classes by learning subtle, discriminative features. In this work, we approach the problem of artist identification as a fine-grained recognition task, where the goal is to distinguish between different artists based on the nuanced visual characteristics of their hatching lines in drawings and prints. Hatching, a popular art technique used to convey tonality, shading, and volume, is often executed quickly and spontaneously, making it a potential carrier of artist-specific, unconscious stylistic signatures. We hypothesize that these subtle variations in hatching patterns encode unique artist-specific features that can be computationally modeled for automated attribution and authentication. To explore this hypothesis, we develop a deep learning-based pipeline capable of detecting hatching regions and learning fine-grained features that discriminate between artists. We evaluate our approach on a diverse collection of drawings and prints spanning multiple artists, artistic styles, and time periods. Our results demonstrate that artist identification from hatching alone is possible, achieving 90-100% accuracy in most cases, highlighting the effectiveness of fine-grained recognition techniques for artist attribution in visual art.