Liu Zhen, Liu Qiang, Gao Gui-Ai, Li Chan
Opt Express. 2017 May 29;25(11):12435-12445. doi: 10.1364/OE.25.012435.
This paper proposes an improved reflectance reconstruction method by adaptively selecting training samples. Modified Principal Component Analysis estimation was proposed by orthogonal regression considering the system noise; deriving the optimum number of training samples by BP-Adaboost neural network; and grouping the representative samples together by hierarchical cluster analysis from a large database of samples. Finally, the training samples were selected by colorimetric subspace tracking. Experimental results indicated that the proposed method significantly outperforms the traditional methods in terms of both spectral and colorimetric accuracy, and our reflectance modeling is a reasonable and convenient tool to generate adaptive training sets.
本文提出了一种通过自适应选择训练样本的改进反射率重建方法。考虑系统噪声，通过正交回归提出了改进的主成分分析估计；利用BP- Adaboost神经网络推导训练样本的最优数量；并从大量样本数据库中通过层次聚类分析将代表性样本分组在一起。最后，通过比色子空间跟踪选择训练样本。实验结果表明，该方法在光谱和比色精度方面均显著优于传统方法，并且我们的反射率建模是生成自适应训练集的合理且便捷的工具。
