基于高光谱图像主成分图像选取与图像融合的彩绘文物图案挖掘

摘要：彩绘文物易碎且珍贵，传统的分析方法已不适用于彩绘文物。针对这一问题，提出了一个新的领域：高光谱成像 技术可以在可见光至近红外光范围内对图像进行观测，且不损伤彩绘文物。本文提出一种自动挖掘彩绘文物图案的方法。 该方法采用最小噪声分率法获取主成分图像，利用显著性目标检测方法获取主成分图像中的显著图像，选取平均梯度值 最大的显著性图像对应的主成分图像作为最优主成分图像，将最优主成分图像与真彩色图像进行融合，得到彩绘文物图 像，从而挖掘出其中的图案。实验数据包括陶器和壁画。结果表明,所提方法能自动、有效地挖掘人眼不易观察到的彩绘 文物图案信息。

关键词：彩绘文物 图案挖掘 高光谱图像 最小噪声分数 显著目标检测 融合

期刊：Journal of Cultural Heritage

DOI：https://DOI.org/10.1016/j.culher.2018.09.008

设备：GaiaField

Mining painted cultural relic patterns based on principal component images selection and image fusion of hyperspectral images

Abstract：Because painted cultural relics are fragile and precious, traditional methods aren’t suitable for analyzing them. A new domain is proposed to resolve this question: hyperspectral imaging technology can observe images when the range of the image is from visible to near infrared light. Further, this can be done without damaging the painted cultural relics. This paper presents a method that can automatically mine painted cultural relic patterns. With this method, the principal component images are obtained via the minimum noise fraction method. A salient object detection method is used to obtain the striking images of the principal component images. The principal component image corresponding to the salient image with the largest average gradient value is selected as the optimal principal component image. The optimal principal component image and true color image are fused, in order to obtain a painted culture relic image so that the patterns are mined. Experimental data includes pottery and murals. The results show that the proposed method can automatically and effectively mine painted cultural relic pattern information that is not easily observed by the human eye.

Keywords：Painted cultural relic Pattern mining Hyperspectral image Minimum noise fraction Salient object detection Fusion