The Extraction And Analysis Of Mural Information Based On Hyper-spectral Data

Mural is the product of history and art. It reflects the economic and culturaldevelopment and ideology of various historical periods in different angles. Theinformation of mural is helpful to help archaeologists deeply understand thehistorical development trajectory and state, analyze the painting techniques of t heearly artists and guide the protection operators to make the reasonable protectionand restoration programs. With the development of hyper-spectral imagingtechnology, it has been widely used in resource, city, ecology, archaeology andother fields. It has many unique advantages which provided a new perspective forthe protection of cultural relics. Hyper-spectral imaging technology has thecharacteristic of wide spectral response range which made it obtain thecomprehensive mural image information. The hyper-spectral image is with a largenumber of bands and narrow band width. These features not only mean theincreasing of the band number, but also make accurate expression of muralinformation possible. The hyper-spectral imaging technology obtains theinformation of the spatial domain which the common image has, in addition, itprovides the information of spectral domain for each pixel which is called―spectrum as image‖. So it can provide a continuous spectral response curve foreach pixel in the image. This provides a basis for the mural information analysisand processing. There had several problems in the information extraction process which were the low extraction efficiency in the underdrawings extraction process,the problem of mixed materials in t he analysis process of mural pigment and theproblem of complex features in the analysis process of mural disease. This papercarried on a thorough research on these problem and got the correspondingsolutions. Specifically, the main research of this paper can be summarized asfollows:（1）. This paper proposed a characteristic band selection method basedon principal component analysis. This method can not only reduce the datadimension, but also remove the redundant information. It’s able to retain the mainunderdrawings information of the mural image by selecting the characteristic band.It would overcome the problem of low extraction efficiency whose reason was thehigh correlation between each band. It’s proved in the experiment that the imagebased on characteristic band not only compressed the data, reduced the duration ofextraction, but also improved the extraction quantity comparing to the originalimage.（2）. This paper proposed a pigment extraction and analysis methodbased on spectral space. Firstly, it est ablished the reference spectral library ofvarious pigments. Secondly, it classified the image to identify the geographicdistribution of each pigment. At last, it would provide accuracy assessment toestimate the classification results. This method effici ently used the hyper-spectraldata’s feature of spectrum as image to determine the area of pigment. It providedguidance for the protection and restoration of mural.（3）. This paper proposed a disease classification method which wasobject-oriented. It analyzed the information of disease such as shape, texture andcolor. It also combined with appropriate classification algorithm in order to achieve effective extraction of disease information. Then it transformed theclassification result to vector image which w as carried on buffer analysis toforecast the possible spread range of disease. It would guide the protection of themural.