Xu Peng, Xu Haisong, Diao Changyu, Ye Zhengnan
Appl Opt. 2017 Oct 20;56(30):8461-8470. doi: 10.1364/AO.56.008461.
A self-training-based spectral reflectance recovery method was developed to accurately reconstruct the spectral images of art paintings with multispectral imaging. By partitioning the multispectral images with the k-means clustering algorithm, the training samples are directly extracted from the art painting itself to restrain the deterioration of spectral estimation caused by the material inconsistency between the training samples and the art painting. Coordinate paper is used to locate the extracted training samples. The spectral reflectances of the extracted training samples are acquired indirectly with a spectroradiometer, and the circle Hough transform is adopted to detect the circle measuring area of the spectroradiometer. Through simulation and a practical experiment, the implementation of the proposed method is explained in detail, and it is verified to have better reflectance recovery performance than that using the commercial target and is comparable to the approach using a painted color target.
一种基于自训练的光谱反射率恢复方法被开发出来，用于通过多光谱成像准确重建艺术绘画的光谱图像。通过使用k均值聚类算法对多光谱图像进行划分，直接从艺术绘画本身提取训练样本，以抑制由于训练样本与艺术绘画之间的材料不一致而导致的光谱估计恶化。使用坐标纸来定位提取的训练样本。利用光谱辐射计间接获取提取的训练样本的光谱反射率，并采用圆形霍夫变换来检测光谱辐射计的圆形测量区域。通过模拟和实际实验，详细说明了所提方法的实现过程，并验证了该方法比使用商业目标具有更好的反射率恢复性能，且与使用彩色绘制目标的方法相当。
