基于流形学习与光谱解混的文物表面字迹增强——以云冈第38窟为例

文物表面的文字是了解历史的关键信息，对文物表面文字信息的探索成为文物保护的重要环节。结合流形学习与光谱解混，提出一种新的文物表面字迹增强方法。首先，利用基于流形学习的等距映射方法对预处理后的高光谱图像进行非线性降维，得到信息量最大的灰度图像；其次，分析文字与背景的光谱特征，通过多层非负矩阵分解方法得到字迹丰度图；然后，将二者进行加权平均得到字迹增强图像，再与合成真彩色影像进行HSV融合，得到字迹融合影像；最后，为更好地辨认文字，在字迹增强图像上裁剪文字并做形态学变换，得到字迹提取图。以云冈石窟第38窟的一景褪色文字高光谱图像为例进行了验证，结果表明，该方法能够有效地增强出文物表面的褪色文字，且较其他增强方法效果更好。

Enhancement of surface handwriting on artifacts based on manifold learning and spectral unmixing:a case study of Cave 38 of Yungang Grottoes

The writing on the surface of artifacts provides key information for understanding history, and the exploration of it becomes an important part of artifact conservation. Combining manifold learning and spectral unmixing, we propose a new method to enhance the writing on the surface of artifacts. Firstly, isometric feature mapping based on manifold learning was used to reduce the nonlinear dimension of the preprocessed hyperspectral images, and a gray image with the most abundant information was obtained. Secondly, the spectral characteristics of text and background were analyzed, and the multilayer nonnegative matrix decomposition method was used to obtain the handwriting abundance map. Then the gray image and the abundance map of the handwriting were weighted together to obtain the handwriting enhancement image; the handwriting fusion image was obtained by HSV fusion of the handwriting enhancement image and true color image. Finally, in order to better identify the text, it was cropped on the handwriting enhancement image and morphological transformation was performed to obtain the handwriting extraction map. The hyperspectral image of a faded text in Cave 38 of Yungang Grottoes was taken as an example, the results showing that the proposed method could effectively enhance the text on the surface of cultural relics and its effect was better than those of other enhancement methods.