摘要
为了去除红外成像非均匀性,提出了一种基于场景自适应方向引导滤波的红外成像非均匀性校正方法。提出的场景自适应方向引导滤波器通过在局部窗口内自适应计算非均匀图像的Canny边缘特征响应,并利用此响应特征计算惩罚因子以调整引导滤波器规整化因子,估计出更真实场景图像,保留图像的细节信息,提高校正算法的鲁棒性。同时,针对神经网络校正算法易产生鬼影的缺点,用运动检测机制以及自适应调整神经网络学习率抑制鬼影。进行了多组红外非均匀性图像校正实验,验证了该红外成像非均匀性校正方法能在有效保留图像细节信息的同时抑制红外像元非均匀性和条纹非均匀性,提高了非均匀校正算法的有效性和鲁棒性。
Infrared imaging non-uniformity severely degrades the quality of infrared images,reducing their clarity and sensitivity,and thereby limiting the effective application of various subsequent infrared image algorithms.To eliminate infrared imaging non-uniformity and improve the quality of infrared images,this paper proposes a novel infrared imaging non-uniformity correction method based on scene-adaptive directional guided filtering,extending the traditional neural network non-uniformity correction algorithm.First,this paper analyzes the causes of non-uniformity in infrared images and proposes a model for infrared non-uniformity generation.It is concluded that infrared non-uniformity manifests as vertical stripe patterns in the vertical direction.Based on this characteristic,the paper improves the Canny extraction algorithm with dual-threshold characteristics and uses the improved Canny algorithm to extract detailed features of infrared images.The dual-threshold characteristic can suppress infrared non-uniformity while extracting detailed image features.This extracted feature is then used to adaptively adjust the regularization factor of the traditional guided filter,resulting in the proposed scene-adaptive directional guided filter.Traditional guided filters process the entire image using uniform linear models and regularization factors,which cannot adequately preserve detailed image information.Although guided filters can protect image edges and textures while smoothing the image,there are still deficiencies when estimating the desired image for the neural network correction algorithm:1)Guided filters use the same regularization factor for all local windows in the image,without considering the differences in detailed textures within different windows.Therefore,the original guided filter does not sufficiently protect the detailed features of the image.2)The uncorrected infrared image contains stripe non-uniformity and pixel response non-uniformity.Although the intensity of these non-uniformities is weaker than that
作者
肖沁
李正周
刘海毅
XIAO Qin;LI Zhengzhou;LIU Haiyi(School of Microelectronics and Communication Engineering,Chongqing University,Chongqing 400044,China)
出处
《光子学报》
EI
CAS
CSCD
北大核心
2024年第11期246-258,共13页
Acta Photonica Sinica
基金
国家自然科学基金(No.61675036)
十三五国防预研基金(No.6140415020312)。
关键词
红外成像
非均匀性校正
引导滤波
神经网络
红外特征提取
场景自适应滤波
Infrared imaging
Non-uniformity correction
Guided filtering
Neural network
Infrared feature extraction
Scene-adaptive filtering
