【作者】 周婷；

【导师】 张小刚； 周浩宇；

【作者基本信息】 湖南大学 ， 控制工程（专业学位）， 2019， 硕士

【摘要】 在雾霾、粉尘环境下采集的图像存在颜色失真、细节缺失等质量问题,在遥感、交通、人脸识别等应用中会给后续处理带来困难,因此图像去雾是机器视觉与图像处理的研究热点,应用面非常广泛。有效提升去雾速度是现有算法的一个关键问题。暗原色先验理论简单有效,基于此理论的去雾算法效果较好而备受关注,但该算法去雾后的图像易受光晕效应影响,需利用抠图算法来准确估计透射率。然而,抠图算法复杂度高且占内存空间大因而不利于实时处理。本文分析雾天图像的降质机制,深入研究了基于GDGIF滤波的图像快速去雾和相关色彩改善方法,并针对回转窑燃煤火焰图像进行了应用研究,在自然图像和工业回转窑燃煤火焰图像上进行了大量的去雾实验。本文研究工作的内容分为以下几点:(1)针对暗原色先验去雾算法中抠图存在耗时严重、占用内存大等问题,研究了基于GDGIF滤波器对透射率进行准确估计的新方法。该算法时间复杂度是线性的,可有效提升算法的效率,且GDGIF滤波器具有出色的边缘保持特性,在景深突变区域仍可精确的估计出透射率,进而能克服黑斑效应、光晕效应。(2)针对去雾算法处理恢复后的去雾图像存在色调暗沉的问题,我们深入剖析并解释了该问题出现的原因,接着重新引入大气散射模型,从该模型出发定义出模拟无雾图,并利用图像融合技术,将该模拟无雾图与基于GDGIF滤波去雾后的去雾图像做像素级融合处理,以改善去雾图像的色彩,解决已有算法的过去雾问题。实验数据表明,该方法可有效提升图像可见度与亮度,且对于在浓雾天气下采集到的带雾图像仍适用。(3)回转窑工业处理过程中采集的燃煤火焰图像因受各类粉尘和光源的影响而模糊不堪,严重影响了该图像后续操作的开展。考虑到已有方法对回转窑火焰图像增强效果不理想,本文提出了一种自适应火焰图像去雾算法,该算法可自动选取大气光值。通过实验证明,本文提出的方法可清晰自然的复原出火焰图像,且图像中的各区域都能得到有效的增强。更多还原

【Abstract】 The image collected in haze and dust environment has some quality problems such as color distortion and lack of details.It will bring difficulties to follow-up processing in remote sensing,traffic,face r ecognition and other applications.Therefore,image defogging is a research hotspot of machine vision and image processing,and its application is very extensive.Effective improvement of defogging speed is a key problem of existing algorithms.The dark channel prior theory is simple and effective,the defogging algorithm based on this theory has a good effect and attracts much attention.But the defogging image is susceptible to halo effect,so matting algorithm is needed to estimate the transmittance accurately.However,matting algorithm has high complexity and occupies large memory space,which is not conducive to real-time processing.In this paper,the mechanism of fog image degradation is analyzed,the fast image de-fogging and related color improvement methods based on GDGIF filtering are studied in depth.Applied research is carried out on the image o f coal-burning flame in rotary kiln.A large number of experiments are carried out on natural image and industrial image of coal-burning flame in rotary kiln.The research work of this paper is divided into the following points:(1)In order to solve the problems of time-consuming and memory-consuming in matting of DCP defogging algorithm,a new method of accurate estimation of transmittance using GDGIF filter is studied.The time complexity of the algorithm is linear,which can effectively improve the efficiency of the algorithm.Moreover,GDGIF filter has excellent edge-preserving characteristics.Especially in the sudden change area of depth of field,it can accurately estimate transmission map.Therefore,the black spot effect and halo effect are effectively overcome.(2)According to the problem of dark tone in the reconstructed defogged image,we have made a thorough analysis and explained the causes of the problem.Then we re-introduce the physical model of atmospheric scattering and define a simulated defogging image.Using image fusion technology,the simulated defogging image is fused with the defogging image based on GDGIF filter at the pixel lev el.Through this method,we can improve the color of defogging image and solve the over-defogging in existing algorithms.The experimental data show that the method can effectively restore the clarity and brightness of the image,and it is still applicable to the foggy images collected in dense fog weather.(3)During the industrial processing of rotary kiln,the kiln flame image is blurred due to the influence of various dust and light sources,which seriously affects the follow-up operation of the image.According to the shortcomings of existing algorithms in enhancing kiln flame image,this paper proposed an adaptive algorithm for defogging the kiln flame image,which can automatically select atmospheric light values.Experiments show that the proposed met hod can restore the kiln flame image clearly and naturally,and all regions in the image can be effectively enhanced.更多还原