基于PCA-Logistic回归模型的图像过曝光区域检测方法  

Image Overexposed Region Detection Method Based on PCALogistic Regression Model

作　　者：陈涛[1] 符均[1] 丁子硬 陈希[1] CHEN Tao;FU Jun;DING Zi-ying;CHEN Xi(Faculty of Electronic and Information Engineering,Xi'an Jiaotong University,Xi’an 710049)

机构地区：[1]西安交通大学电子与信息学部,陕西西安710049

出　　处：《制造业自动化》2025年第4期40-47,共8页Manufacturing Automation

基　　金：科技创新2030——“新一代人工智能”重大项目(2022ZD0115803)。

摘　　要：针对过曝光区域检测问题,提出了一种基于主成分分析(Principal Components Analysis,PCA)和Logistic回归的过曝光图像饱和像素检测方法。首先通过研究分析过曝光图像的显著性特征,提取了亮度及颜色特征、人类视觉修正的饱和度特征、空间邻域特征、局部熵特征、灰度对比度特征等变量作为检测图像过曝光的初始指标;接着利用主成分分析方法对原始指标变量进行降维处理,然后利用建立的L2正则化的Logistic回归模型进行分析预测;最后与其他过曝光检测算法进行了对比分析,并在某安防监控图像中进行了过曝光区域检测效果验证。结果表明,该模型检测结果更具整体性,检测区域更紧凑,也更符合人眼对过曝光区域的视觉感知。For the need of detecting overexposed regions,a saturated pixel detection method for overexposed images based on Principal Component Analysis(PCA)and Logistic Regression is proposed.By analyzing the salient features of overexposed images,the variables such as brightness and color features,saturation features corrected by human vision,spatial neighborhood features,local entropy features,and grayscale contrast features are selected as the initial indexes for detecting overexposure of images;the original index variables are downscaled by using principal component analysis,and then analyzed by using the established logistic regression model with L2 regularization.Finally,it is compared and analyzed with other overexposure detection algorithms,and the effect of overexposed region detection is verified in a security monitoring image.The results show that the detection results of this model are more holistic and the detection area is more compact,which is more in line with human's visual perception of overexposed areas.

关 键 词：过曝光图像 饱和像素检测 主成分分析(PCA) LOGISTIC回归分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]