计算机应用研究2024，Vol.41Issue(10)：3188-3193,6.DOI:10.19734/j.issn.1001-3695.2023.11.0597
残差修正的加权多项式回归色彩特征化算法
Weighted polynomial regression color characterization algorithm with residual correction
摘要
Abstract
In the field of digital printing,accurately reproducing the color of computer images is a prerequisite for high-quality printing,where color characterization is a key step.Traditional polynomial regression models tend to amplify outliers in the characterization sample set due to high-order terms,causing model oscillation and affecting the accuracy of color characteriza-tion.Color characterization algorithms based on neural network have higher precision but significantly increase in algorithmic complexity,making them unsuitable for the efficiency requirements in industrial production.To address these issues,this pa-per proposed a color characterization method based on weighted polynomial regression algorithm with residual correction.This algorithm employed the Huber loss function,known for its strong robustness against outliers,as a substitute for mean squared error.It determined the weight of each sample through an adaptive mechanism and iteratively optimizes the residual values to obtain the optimal weight matrix,effectively reducing the impact of outlier samples on the system.Additionally,the correction module captured nonlinear scenarios that the initial model might miss,significantly improving the adjustment of the transforma-tion results and thereby enhanced characterization precision.The results show that compared to conventional polynomial regres-sion,this algorithm reduces the average color difference by 1.2.It achieves a precision close to that of deep belief network al-gorithms but with more than 99.37％reduction in inference time.关键词
色彩特征化/多项式回归/自适应加权/色彩复制/色彩管理Key words
color characterization/polynomial regression/adaptive weighted/color reproduction/color management分类
信息技术与安全科学引用本文复制引用
杨晨,廉凯成,徐昊,吴秦,柴志雷..残差修正的加权多项式回归色彩特征化算法[J].计算机应用研究,2024,41(10):3188-3193,6.基金项目
国家自然科学基金资助项目(61972180) （61972180）
江苏省模式识别与计算智能工程实验室资助项目 （）
