基于结构风险最小化和全最小二乘法的色彩校正

该文提出了一种局部回归色彩校正算法。该算法从结构风险最小化原理出发,提出以全最小二乘法残差为经验风险,选取校正颜色点的K近邻域,采用局部回归进行色彩校正。实验结果表明该文算法在精度和鲁棒性方面均优于多重回归校正以及分区回归校正,其平均误差、最大误差,误差标准差分别下降了46%(27%),57%(21%),42%(20%)。

Color Calibration Based on Structural Risk Minimization and Total Least Squares

A locally regressive algorithm for color calibration is proposed. Starting from the principle of structural risk minimization, the algorithm regards the residual of total least squares as the empirical risk and chooses the K-nearest neighborhood of a calibration color point to implement local regression for color calibration. Experimental results indicate that the proposed algorithm is superior, in both precision and robustness, to multiple regression and multiple regression based on subspaces and that its average error, maximum error and error standard deviation decrease by 46%(27%), 57%(21%) and 42%(20%) respectively.