Deterioration Extraction for Ancient Murals Via a Decision Tree with a Deterioration Index

Mural painting is the earliest independent form of painting and involves the crystallization of ancient traditional painting arts. However, over time, various types of deterioration have influenced murals, among which the most common is the deterioration of the plaster layer. Existing extraction methods largely use the shape features obtained from digital images, which are influenced by the diversity of information on the murals. To overcome this drawback, hyperspectral technology was introduced to achieve accurate deterioration extraction. First, the spectral features of deterioration, with a focus on the plaster layer, were analyzed. The decision tree deterioration index was subsequently constructed to extract deterioration information. Combining the threshold of the near-infrared reflectance and the difference between the reflectance at 440 and 490 nm, our method was effective for complete extraction. The method achieved better visual results than other traditional extraction methods did. The precision, recall, F-measure, and overall accuracy were 0.7244, 0.7581, 0.7409, and 0.9680, respectively. Our method also has high threshold stability, yielding good results for other mural images.