基于空谱特征联合的烟熏壁画线条增强方法

Line enhancement method of smoky murals based on combination of spatial spectrum features

针对壁画部分区域被烟熏污染,人眼难以辨别的问题,现有的图像增强算法未能充分考虑壁画影像中的同质区域和空谱信息,因此,本文提出了基于空谱特征联合的烟熏壁画线条增强方法。首先,用地面高光谱成像仪采集壁画的原始高光谱数据进行数据预处理,选取波段合成真彩色影像并利用多尺度模糊C均值算法进行图像分类,将分类后数据与高光谱影像一一对应得到分区高光谱数据,对各分区数据进行核主成分分析(kernel principal componentanalysis,KPCA)降维获取前几主成分,经平均梯度分析选出分区KPCA的最优主成分影像;其次,分析壁画每幅影像的行列相关性,对预处理后的高光谱影像进行二维主成分分析获取其行、列主成分,经图像重构和平均梯度分析获得二维主成分分析的最优主成分影像;最后,结合光谱特征分析与自适应伽马校正算法得到包含壁画线条增强信息和色彩信息的影像。结果表明,与已有的单尺度Retinex算法、带色彩恢复的多尺度Retinex算法及侯妙乐等(2014)方法进行对比,本方法得到的影像线条完整、清晰,具有更好的对比度。

Murals are the carrier of our country’s history and the main component of cultural heritage,with extremely high historical and artistic value.However,due to the long-term influence of natural or man-made factors,the murals’partial lines may be covered by smoke or other substances.This significantly affects the murals’appearance and is not conducive to their digital preservation.It is urgent to carry out a scientific restoration of murals and formulate a reasonable and scientific digital protection plan for murals,and the enhancement and extraction of mural line information is one of the indispensable components.The advantages of hyperspectral imaging technology were used in this work to analyze the spectral and spatial characteristics of lines as well as various color pattern information on murals.The smoked fresco line enhancement method,which is based on the combination of spatial-spectral features,was proposed along with the dimensionality reduction algorithm and image unmixing method.Initially,reflectance correction and image denoising for the mural hyperspectral data are the two key components of the data preprocessing that is performed on the original hyperspectral images of the mural target area.Secondly,the multi-scale fuzzy C-means algorithm is used to classify the synthetic true color images.Local spatial information present in the classified data reduces the influence of noise and artificial influences.The classified data is divided into multiple homogeneous regions andcorresponds point by point to the mural hyperspectral data.Perform the KPCA algorithm on each homogeneous subregion to obtain a dimensionality reduction matrix.After image reconstruction,a principal component image is obtained,and the average gradient information of each principal component image is calculated to screen out the principal component image with the most abundant line information,and fully exploit the nonlinearity in the mural line feature.The hyperspectral images of murals are rotated and transformed to obtain the rotated images,which are then subjected to 2DPCA transformation to obtain the row and column principal component images.This process takes into account the variations in rows and columns in different hyperspectral images.The row and column principal components are arranged correspondingly.The weighted average method is used for image fusion to obtain the final principal component image and the average gradient information of each principal component image is calculated to screen out the principal component image with the most abundant line information.VCA is used to extract endmembers from hyperspectral images in conjunction with the analysis of the spectral characteristics of murals.The non-negative constrained least squares algorithm is used to obtain the abundance map of each endmember,and the average gradient of the corresponding abundance map of each endmember is calculated to select line information.The most abundant images are the line abundance maps.After normalizing the optimal principal component images of the subregional KPCA and 2DPCA,the line abundance maps with richer line information are weighted and summed to obtain the line information enhanced image.The line information enhanced images are fused with the synthetic true color image to obtain the line information.The information fusion image has a certain color difference between the fused image and the digital image.Therefore,the brightness component is extracted by converting the color space,and the adaptive gamma correction and weighted distribution function smoothing are performed on it to obtain the final image,which realizes the enhancement of mural line information.The qualitative and quantitative analysis of images shows that the lines in images obtained by this method are significantly enhanced compared with original grayscale images and synthetic true color images,and the images contain more line information.Compared with various image enhancement algorithms such as SSR,MSRCR,and mural manuscript hyperspectral information enhancement methods,the proposed method can better distinguish mural lines from other pattern backgrounds and enhance the contrast between lines and other patterns.Compared with digital images,the proposed method can better restore the color of the murals,making the murals more layered and the characters more vivid.It shows that this method can use hyperspectral data to enhance the line information in the area covered by the material in the mural,effectively improve the contrast between the mural lines and other patterns,and improve the visual effect of the mural.