Abstract: Abstract: The extraction of mural paint loss plays an important role in investigating the present situation of murals. Given the similarity of spectral features of paint loss to the white patterns of the mural, the only utilization of spectral features would make it less accurate for the extraction of the mural paint loss. Aiming at improving the extraction performance, a comprehensive method was proposed that integrated spectral features and Hu moment. First, the supervised support vector machine method was employed to classify paint loss and white patterns by the spectral features which the dimension had been reduced. The classified pixels of paint loss and white patterns were then connected to form two types of polygons, which were regarded as the smallest indivisible objects. Subsequently, seven Hu moments for each polygon were calculated as the shape features serving as the feature vectors to distinguish each polygon again based on the support vector machine method. In this way, the semi-automatic extraction of mural paint loss was realized. A case study was conducted to evaluate the performance of our proposed method, using the hyperspectral data of the Qutan Temple mural. The results show that our proposed method is capable of enhancing the extraction accuracy of mural paint loss and supporting the investigation of the present situation of murals.

Key words: Keywords: paint loss, disease extraction, mural, hyperspectral imaging, Hu moment, support vector machine, minimum noise fraction