Detection of alterations in historical violins with optical monitoring

Preventive conservation is the constant monitoring of the state of conservation of an artwork to reduce the risk of damage in order to minimise the necessity of restorations. Many methods have been proposed to achieve this goal, generally including a mix of different analytical techniques. In this work, we present two probabilistic clustering algorithms for the detection of alterations on varnished surfaces, in particular those of historical musical instruments. Both methods are based on the a-contrario framework and the Number of False Alarms (NFA) criterion. The first one tackles the problem of detecting changes between a pair of colour images by analysing their difference map. It considers simultaneously grey-level and spatial density information with a single background model. The second method works with a sequence of images and analyses the evolution of the changed areas between frames. Both methods are robust to noise and avoid parameter tuning as well as any assumption about the shape and size of the changed areas. In both cases, tests have been conducted on UV-induced fluorescence (UVIFL) image sequences included in the “Violins UVIFL imagery” dataset. UVIFL photography is a well-known diagnostic technique used to see details of a surface not perceivable with visible light. The obtained results prove the capability of the algorithm to properly detect the altered regions. Comparisons with other state-of-the-art clustering methods show improvement in both precision and recall.