[1A4] Extending B-COSFIRE for automatic extraction of craquelure
M Fernandes¹,², R Araújo¹, G Almeida¹ and S Paredes²
¹JTA, The Data Scientists, Portugal
²Polytechnic Institute of Coimbra, Portugal
This study addresses the segmentation of craquelure patterns in fine arts, a complex challenge due to the varied textures, colour transitions and canvas deformations typical in paintings. We advance this field by introducing supervised learning, manually labelling two distinct image datasets: one comprising greyscale images from the established Bucklow’s dataset; and the other consisting of image patches extracted from historical paintings, reflecting more realistic scenarios. We employ the B-COSFIRE model, originally developed for medical imaging, to enhance the detection and delineation of these complex patterns. This approach significantly boosts performance by optimising the model’s parameters using our labelled datasets. We qualitatively compare the B-COSFIRE model’s effectiveness against a state-of-the-art segmentation model. Our findings demonstrate that supervised learning significantly improves the ability of the model ability to accurately identify and segment craquelure, underscoring its potential in the domain of fine art analysis.
Keywords: curvilinear structure segmentation, craquelure extraction, computer vision, supervised learning, fine arts.
Keywords: curvilinear structure segmentation, craquelure extraction, computer vision, supervised learning, fine arts.
