Domain-Informed and Neural-Optimized Belief Assignments: A Framework Applied to Cultural Heritage

Abstract

Identifying pigments in Cultural Heritage artifacts is key to uncovering their origin and guiding conservation strategies. Although recent advances in non-invasive imaging have enabled the collection of rich multimodal data, existing methods often fall short in dealing with uncertain, ambiguous, or noisy information. This paper introduces a versatile fusion framework based on Belief Function Theory, combining expert-driven evidence modeling with neural optimization. Our main contribution lies in a general strategy for assigning mass functions that captures prior knowledge and adapts through task-specific training. When applied to pigment classification, our method demonstrates robustness against source variability and ambiguity. Experiments conducted on both synthetic and mock-up datasets validate its effectiveness and suggest promising potential for broader applications.

Keywords: Belief Function Theory, Mass Function Assignment, Neural Network-based Optimization, Cultural
Heritage, Pigment Identification