【作者】 杨伟；

【导师】 程建；

【作者基本信息】 电子科技大学 ， 工程硕士（专业学位）， 2017， 硕士

【摘要】 近年来,我国很多地区都遭受到了不同程度的雾霾影响,这给我们的日常生活带来了诸多的不便,尤其是体现在智能交通、航天航空领域中。正是由于雾霾或者大雾的存在,我们在户外拍摄的图像往往表现出质量下降严重,表现为对比度不足、颜色模糊等现象,这不仅不利于我们人眼的视觉体验,而且对我们提取图像中的相关特征带来了很大的挑战。因此,关于有雾图像的去雾方法一直是计算机视觉领域的一大研究热点,并且具有广泛的应用前景。本文针对自然场景图像去雾方法研究,主要研究内容如下:首先,本文介绍了大气散射原理,并着重地介绍了大气散射模型的推导和建立,并对有雾图像的基本特性进行了分析,这对接下来的有雾图像去雾的方法研究是个很好的铺垫。其次,研究了两种经典的基于物理模型的有雾图像复原方法,分别是基于大气光幕的去雾方法和基于优化对比度增强的去雾方法。在介绍前者的时候分析了该方法的两个不足之处:即参数不易调整和中值滤波保边缘性不算太强。而在介绍后者时,充分地突出了该方法的有效性和稳定性,然后分别从主观人眼视觉和客观定量分析定量两种不同的角度来对这两种去雾方法的复原情况进行评估,并且对本文的方法和传统的经典方法结果进行比较。实验表明,本文的方法速度上能到达实时效果,而且复原后的图像质量不论是边缘细节还是颜色效果相对于传统方法都有较大的提高。最后,介绍了图像去雾相关领域里著名的基于暗通道先验理论的图像去雾方法。详细地论述了该方法的优势及原理。分析了该方法的缺点,并针对它的缺点在此基础上提出了一种改进的暗通道先验的去雾方法。该方法用导向滤波代替soft matting,让速度得到了较大的改善。并针对He方法对天空区域的失效性增加了处理天空部分的操作,接着定性分析和定量评估了该方法与经典的去雾方法之间的去雾性能。实验证明该方法具有很好的速度,能解决很多经典去雾方法克服不了的问题。更多还原

【Abstract】 In recent years,many parts of our country have been affected by the varying degrees of haze,which makes our daily life inconvenience,especially in the intelligent transportation,and aerospace field.Because of the existence of fog or haze,the quality of the images captured in the outdoor has some serious decline,such as lack of contrast,color blur and so on,which is not only disadvantageous for the visual experience of our eyes,but also makes the image features extraction great challenging.Therefore,the defogging of fog images have always been a hot topic in computer vision,and has wide application prospect.In this thesis,the defogging method of the natural scene image was studied,and the main research contents as follows:Firstly,the principle of atmospheric scattering was introduced,especially the derivation and establishment of the atmospheric scattering model.The basic characteristics of the fog image was analyzed,which establishes a theoretical foundation for defogging study.Secondly,two kinds of foggy image restoration methods with physical models were studied,which are based on atmospheric light curtain defogging method and based on optimized contrast enhancement defogging method respectively.The two shortcomings of the former method were analyzed: the parameters are not easy to adjust and the median filter can not keep the edges strong.For the latter method,the effectiveness and the stability were fully highlighted.The performance of image restoration of the two defogging methods were evaluated from two different perspectives,such as the subjective human vision and the objective quantitative analysis.Moreover,these two proposed methods were compared with the traditional classical methods.Experimental results show that the computation speed of two proposed methods can reach the real-time,and the image qualities after restoration,such as the edge of the details and the color fidelity,have been greatly improved.Finally,the famous image defogging method based on the dark channel prior theory was introduced.The advantages and the principles of this method were discussed in detail.Then,the shortcomings of this method was analyzed,and an improved defogging method was proposed based on the dark channel prior theory.The soft matting is replaced by the guided filtering in the proposed method,so that the computation speed has been greatly improved.Moreover,for the failure of the He’s method defogging the sky area,the operation on the sky area defogging was proposed.Then qualitative analysis and quantitative assessment was made between the proposed method and the classical methods.Experiments show that the proposed method has a very good computation speed,and can solve many problems in defogging which cannot be overcome by the classical methods.更多还原