摘要
壁画是人类文明的瑰宝,具有极高的历史意义与艺术价值,而颜料是壁画表面信息的重要组成部分.针对壁画表面同色系颜料肉眼难以辨认的问题,提出了基于归一化光谱指数的颜料识别方法.利用地物波谱仪获取颜料的光谱反射率,覆盖波长为350～2 500 nm.通过一阶导数初步选择特征波段,利用欧氏距离作为可分性指标确定最终4个特征波段,构建归一化光谱特征指数.通过区分度来选取最优指数,区分度大于0.5则认为颜料易于区分.以绿色系颜料为例,选取氯铜矿、铜绿、绿玛瑙、绿松石、群绿、石绿、云母绿7种绿色颜料,构建了绿色系不同颜料的光谱特征指数,并以同样的方法构建了其他色系的光谱特征指数.以模拟壁画与云冈石窟06窟西壁壁画作为验证数据,利用所提出的同色系光谱特征指数、光谱角、欧氏距离、光谱特征拟合、二进制编码匹配分别进行颜料识别.结果表明所提方法能有效识别同色系颜料,对于壁画同色系颜料的快速准确识别具有实践意义.
Abstract
Mural paintings and other painted cultural relics are precious cultural heritage with great artistic value and historical significance.The pigment is an important part of the information on the surface of a mural painting,and it is extremely important to identify the pigment type quickly and accurately.Due to the non-renewable nature of cultural relics,the use of non-destructive means of detection has become necessary.The hyperspectral technology has become an important research tool for identifying mural pigments due to its advantages of non-destructive detection,high spectral resolution,as well as the union of imagery and spectrum.In this study,a pigment identification method using Euclidean distance(ED)combined with the normalized spectral characteristic index(NDSI)is proposed to address the problem of differentiating pigments with the same color in murals.Firstly,seven green mineral pigments,including atacamite,verdigris,green agate,turquoise,ultramarine green,malachite,and mica green,were selected and painted on rice paper to form samples.We obtained their reflectance in the wavelength of 350-2 500 nm using an ASD FieldSpec4 spectroradiometer.Secondly,the spectral first-order derivatives were utilized in the selection of characteristic bands,which can highlight the curve trend.In this case,the bands with a derivative value of 0 are the peaks and troughs of a curve,which can be considered as characteristic bands.Taking the atacamite pigment as the target pigment,for the i-th characteristic band of atacamite pigment,we calculated the sum of the Euclidean distances between atacamite and other pigment spectra in this band to obtain the distance index Di corresponding to the i-th characteristic band.The distance indices for all characteristic bands were sorted in a descending order,and the four bands corresponding to the top four largest distance indices were selected as the final characteristic bands of the target pigment.Then,the NDSI was constructed using these characteristic bands,which are given by NDSI=1/2(Ra-Rb/Ra+Rb+Rc-Rd/Rc+Rd),where Ra,Rb,Rc and Rd are the reflectance values of the characteristic bands respectively.Fifteen NDSI values can be obtained by permutations and combinations between bands.In order to select the optimal NDSI,the distinction between the 15 indices was calculated,which is given by Q=|q-W/q|,Where q is the NDSI value of the target pigment and W is the index value calculated by substituting the NDSI formula of the target pigment into the spectra of other pigments.We assume that indices greater than 0.5 can distinguish the target pigment from other pigments.We have simultaneously applied the maximum-minimum selection method here.When the minimum distinction of a certain index for all pigments is the highest among all indices and the distinction is greater than 0.5,the index is considered as the optimal NDSI for the target pigment.Finally,NDSIs were constructed for all green pigments using the above steps.In addition,we selected optimal NDSIs for red,yellow and blue pigments at the same time in the experiments.In order to verify the accuracy of the spectral feature indices selected for the experiments,a laboratory simulated mural was firstly used as the validation data.The spectra of pigments at seven points on the surface of the simulated mural were collected for pigment identification using NDSI,spectral angle mapping(SAM),ED,spectral feature fitting(SFF),and binary encoding(BE),respectively.The results show that the method was capable of distinguishing between pigments of the same color category compared to other spectral matching methods,improving the accuracy of pigment identification.In order to verify the applicability of NDSI on real mural artefacts,the pigments on the surface of the mural paintings on the east wall of Cave 6 of Yungang Grottoes,Shanxi Province,China were selected as the validation data.By using NDSI,we found that the pigment of the blue area was lazurite,and the pigments of three red areas were cinnabar for one and red lead for two.The identification results were basically consistent with the detection results of Raman spectrometry.In summary,the pigment identification method proposed in this paper for pigments of the same color category could quickly extract the pigment feature information and carry out pigment identification with higher accuracy,which is of practical significance for the identification and differentiation of pigments of the same color category in murals.However,it is hard to identify mixed pigments using NDSI,a problem that needs to be improved upon future studies.
维普
万方数据