基于提升核偏最小二乘回归的色彩校正

Color calibration based on boosting kernel partial least squares regression

为进一步提高回归算法的色彩校正精度,提出一个基于核偏最小二乘回归的局部迭代算法。该算法首先将源色彩空间中校正样本的邻域样本经核函数映射到一个高维的线性特征空间,然后在提升的迭代过程中通过偏最小二乘回归消除映射数据的多重共线性,最终达到色彩校正的目的。实验证明算法引入的特征空间提供了额外的校正信息,提升方法增强了核偏最小二乘回归性能,算法在精度和鲁棒性上均优于传统的多重回归和基于分区的回归校正方法。

To further improve the calibration precision of the previous regression algorithms, an iterative local algorithm of color calibration based on kernel partial least squares regression is proposed. The algorithm firstly maps the local input data of the calibration color sample in source color space into a linear feature space via kernels, and then eliminates the muhi-linearity among the mapped data using partial least squares regression in each boosting iteration so as to realize color calibration. Experimental results indicate that the feature space provides extra calibration information, boosting enhances the regression performance of kernel partial least squares and the proposed algorithm turns out to be superior in both precision and robustness to multiple regression and its variant based on subspaces.