摘要
针对现有去雾方法在含有大面积亮白区域的图像中,传统方法有可能导致大气光值估计不准确,提出基于四叉树分解的方法,避免亮白区域对大气光估计的影响,在天空区域内对大气光进行准确估计;同时,为避免手工特征提取及假设条件的限制,利用三个不同尺度的卷积核对原始雾图进行卷积操作,经过网络的一系列特征学习之后得到待细化传播图;然后使用图像融合方法对其进行细化;最后,将估计的参数代入大气散射模型从而反演出清晰图像。合成和真实世界的粮库雾尘图像的定量和定性实验结果表明,该算法对于图像纹理细节以及天空区域的处理上有较好效果,且鲁棒性高,普适性强。
As the traditional defogging method could cause the inaccurate estimation of atmospheric light value in the images containing large areas of bright white,a method based on Quadtree decomposition is proposed,which avoids the effect of bright white areas on the atmospheric light estimate,and accurately estimates the atmosphere light in the sky area.In order to avoid the limitation of handwork feature extraction and assumed condition,the convolution operation of the original fog map is conducted by means of the three different⁃scale convolution kernels,the diffusion map to be refined is obtained by a series of feature learning of the network,and then the image fusion method is used to refine it.The estimated parameters are incorporated into the atmospheric scattering model to produce a clear image.The experimental results of quantitative and qualitative of the synthetic and real⁃world grain depot fog image demonstrate that the algorithm has better effect on image texture details and the processing of sky area,and has high robustness and strong universality.
作者
桂便
祝玉华
甄彤
GUI Bian;ZHU Yuhua;ZHEN Tong(College of Information Science and Engineering,Henan University of Technology,Zhengzhou 450001,China)
出处
《现代电子技术》
北大核心
2020年第14期121-124,127,共5页
Modern Electronics Technique
基金
国家重点研发项目:“北粮南运”散粮集装箱高效保质运输技术及物流信息追溯平台支撑示范工程(2018YFD0401404)
国家重点研发项目:新型粮情测控技术与装备开发(2017YFD0401004)。
关键词
粮库图像
图像去雾
四叉树分解
光值估计
卷积网络
图像融合
grain image
image fogging
Quadtree decomposition
light value estimation
convolution network
image fusion
