弱光照图像的模糊细节自调节增强测试与仿真  

Testing and Simulation of Fuzzy Detail Self adjustment Enhancement in Weak Illumination Images

作　　者：张瑾[1] 吴立春[2] ZHANG Jin;WU Li-chun(Yinchuan University of Energy,Yinchuan Ningxia 750100,China;School of Medicine and Information Engineering,Ningxia Medical University,Yinchuan Ningxia 750004,China)

机构地区：[1]银川能源学院,宁夏银川750100 [2]宁夏医科大学医学信息与工程学院,宁夏银川750004

出　　处：《计算机仿真》2024年第6期308-312,共5页Computer Simulation

基　　金：银川能源学院学科专业带头人及优秀骨干教师培养项目(银能校【2022】274);2020年宁夏回族自治区高校一流本科专业建设项目(宁教高办[2021]7)。

摘　　要：由于自然光具有较强的随机性,低光照条件下相机采集的图像较为模糊,难以获取细节信息。为此,提出弱光照图像模糊细节自调节增强方法。在非局部均衡滤波算法中引入相似度阈值函数,去除弱光照模糊图像噪声,采用分数阶全变差模型和非局部全变差模型,建立弱光照模糊图像去模糊模型,对模糊图像去模糊处理,将预处理后弱光照图像输入多尺度融合卷积神经网络模型,通过全局路径和局部路径提取并融合图像特征,实现弱光照图像模糊细节自调节增强。实验结果表明,所提方法在合成图像增强中均方误差更小、结构相似度和峰值信噪比更高,在真实图像增强中自然图像质量评价度量和对比度增益更高、熵值更接近于8,图像处理时间更短,仅为0.745ms。Due to the strong randomness of natural light,the images captured by cameras under weak light are of⁃ten blurry,and it is also difficult to obtain detail information.To address this problem,this paper presented a self-reg⁃ulating enhancement method for blurred details in weak lit images.At first,a similarity threshold function was intro⁃duced into the non-local equalization filtering algorithm to remove noise from blurred images.Then,the fractional or⁃der total variation model and the non-local total variation model were used to construct a deblurring model.After the image deblurring,the preprocessed images were input into a multi-scale convolutional neural network model.Finally,the image features were extracted and fused through global and local paths,thereby achieving self-regulating enhance⁃ment of blurred details in weak-illumination images.The experimental results show that the proposed method has smaller mean square error,higher structural similarity and peak signal-to-noise ratio in synthetic image enhancement,while in real image enhancement,it has higher evaluation metric of natural image quality and higher contrast gain.Meanwhile,the entropy is closer to 8,and the image processing time is only 0.745ms.

关 键 词：弱光照图像 模糊细节 自调节增强 非局部均值滤波 卷积神经网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]