Virtual restoration of stains on ancient paintings with maximum noise fraction transformation based on the hyperspectral imaging

Abstract Ancient paintings, as one of the most important forms of artistic expression of Chinese traditional culture, are the most valuable and non-renewable treasure of human civilization. However, unfortunate situations occur, causing stains on paintings. Stains disfigure their artistry and values, and it is desirable to remove them. Traditional removal methods using physical means or chemicals may damage the original paintings. Recent virtual restoration effort may cause inconsistent content when applied to larger regions. This paper proposes a new virtual restoration method of stains based on the maximum noise fraction (MNF) transformation with the hyperspectral imaging. The method has two steps. Firstly, it carries out the forward MNF transformation to concentrate the main features of ancient paintings into the several top principal components. Secondly, it determines the principal component that contains the large spectral information of stains, and applies the inverse MNF transformation to several top components except for the chosen components to reduce the stain effect on the image and restore the original spectral information and color as much as possible. This paper selects a paper painting of the Qing Dynasty as the experiment data, and the results show that the method has the effect of diluting or eliminating image spots, and can restore the style of ancient paintings to a large extent without causing a large loss of data information.