摘要
在智慧交通领域中,监控拍摄的车辆图像质量会受到场景光照条件的影响而造成关键信息损失,导致后续交通执法缺乏可靠的图像依据,而现有的图像曝光校正方法存在校正后图像颜色失真、细节模糊等问题.为此,提出一种面向智慧交通的双重特征融合图像曝光校正方法.该方法通过全局子网与曝光子网提取相关特征,全局子网引入特征注意力机制提取图像全局特征,改善校正图像的整体亮度与颜色;曝光子网基于U-Net结构,引入多尺度注意力机制提取图像中不同尺度的曝光特征,提升校正图像的细节纹理;最后对双重特征进行融合,将其映射至图像空间,实现图像曝光校正.在MSEC数据集上的实验结果表明,所提方法的峰值信噪比达到21.801 5 dB,结构相似度达到0.867 7,自然图像质量评价达到11.711 8,均优于对比的曝光校正方法,能够有效地还原图像颜色信息与细节纹理,证明该方法具有良好的曝光校正效果.
In the field of intelligent transportation,the quality of vehicle images captured by surveillance cameras can be affected by the lighting conditions of the scene,resulting in the loss of critical information,which in turn leads to a lack of reliable image evidence for subsequent traffic enforcement.However,existing image exposure correction methods have problems such as color distortion and blurry details in the corrected image.Therefore,a dual feature fusion image exposure correction method for intelligent transportation is proposed,which extracts relevant features through global subnet and exposure subnet,respectively.The global subnet introduces the feature attention mechanism to extract global features of the image,improving the overall brightness and color of the corrected image.The exposure subnet is based on the U-Net structure and introduces a multi-scale attention mechanism to extract exposure features at different scales in the image,improving the detailed texture of the corrected image.Finally,image exposure correction is achieved by fusing the dual features and mapping them to the image space.The experimental results on the MSEC dataset demonstrate that,compared to the existing exposure correction methods,the proposed method has a peak signal-to-noise ratio of 21.8015 dB,a structural similarity index measure of 0.8677,and a natural image quality evaluator of 11.7118.The proposed method can effectively restore color information and detailed textures in images,proving its excellent exposure correction effect.
作者
杨佳钧
张金艺
陈琪
Yang Jiajun;Zhang Jinyi;Chen Qi(Key Laboratory of Specialty Fiber Optics and Optical Access Networks,Shanghai University,Shanghai 200444;Joint International Research Laboratory of Specialty Fiber Optics and Advanced Communication,Shanghai University,Shanghai 200444;Shanghai Sansi Institute for System Integration,Shanghai 201100)
出处
《计算机辅助设计与图形学学报》
北大核心
2025年第4期678-687,共10页
Journal of Computer-Aided Design & Computer Graphics
基金
高等学校学科创新引智计划(111)项目(D20031)
上海市闵行区重大产业技术攻关计划(2022MH-ZD19)。
关键词
图像曝光校正
注意力机制
双重特征
智慧交通
image exposure correction
attention mechanism
dual feature
intelligent transportation
