单幅雾天图像去雾算法研究

发布时间：2018-12-13 09:44

【摘要】：近年来,雾霾一次又一次频繁的出现在大家的视野之内,不仅影响到人们的身体健康,同时严重危害到人们的公共交通安全。针对雾天天气情况下获取的各种图像进行去雾研究,本文分析了几种基于图像增强的去雾算法在去雾领域的不足之处。同时,在前人对基于模型的去雾的研究以及在进行理论分析的基础之上,提出透射率趋势图去雾的概念,并针对传统的暗原色先验理论去雾存在的问题提出几点改进想法。(1)针对传统暗原色明亮区域恢复后颜色失真问题,提出明亮区域透射率补偿方法进行修正。通过引入明亮区判别阈值,然后根据原始图像像素亮度值与大气光值差的绝对值与阈值的大小作为判别依据,比阈值小则为明亮区,然后通过透射率修正函数进行修正;反之为非明亮区,则保持原有的透射率。实验证明本文所提透射率修正方法有效解决了明亮区颜色失真现象;(2)针对传统暗原色恢复结果亮度偏暗以及过饱和现象。通过分析大气散射理论并结合成像设备的局限性提出采用透射率趋势图进行去雾的概念。文中采用对粗透射率图进行空间平滑滤波的思想来获取透射率趋势图。实验证明,利用透射率趋势图具有较好的去雾效果,改善了原始算法去雾结果亮度偏暗以及过度饱和的问题;(3)针对原始算法效率低下的问题,本文结合透射率趋势图的概念,提出两种粗透射率图滤波平滑的改进方法:其一,基于高斯平滑滤波的概念,提出基于改进的高斯空间平滑滤波处理来获取透射率趋势图;其二,基于改进的引导滤波的方法,通过重构引导矩阵对粗透射率图平滑处理,不仅能够获得等价细腻的透射率趋势图,同时降低了传统引导滤波对透射率优化的时间开销,其算法的执行效率为原始算法的3~4倍;(4)基于可见边增强的去雾质量评价方法,通过引入色彩信息恢复度指标,结合清晰度指标,构建无参考的图像去雾质量综合评价指标,更加全面的对去雾结果进行评价,论文基于Matlab平台,编写相关算法程序,对上述算法进行实验仿真。实验结果表明,本文提出的改进方案有效的改善和解决了传统暗原色先验去雾算法存在的问题。
[Abstract]:In recent years, haze appears frequently in the field of vision, which not only affects people's health, but also seriously endangers people's public transportation safety. In this paper, we analyze the shortcomings of several image enhancement algorithms in the field of fog removal. At the same time, on the basis of previous researches on model-based de-fogging and theoretical analysis, the concept of transmittance trend map de-fogging is proposed. In order to solve the problem of removing fog in the traditional priori theory of dark primary color, some improvement ideas are put forward. (1) aiming at the problem of color distortion after restoration of the bright region of traditional dark primary color, a method of compensation for the transmittance of bright region is proposed. By introducing the bright region to distinguish the threshold value, then according to the original image pixel brightness value and atmospheric light value difference between the absolute value and the size of the threshold as the basis, smaller than the threshold value is the bright region, and then through the transmittance correction function to correct; On the contrary, if the region is not bright, the original transmittance will be maintained. Experiments show that the proposed transmittance correction method can effectively solve the bright color distortion phenomenon. (2) aiming at the traditional dark primary color restoration results brightness dim and supersaturation phenomenon. Based on the analysis of atmospheric scattering theory and the limitation of imaging equipment, the concept of defogging using transmittance trend map is proposed. In this paper, the idea of spatial smoothing filter for coarse transmittance map is used to obtain the transmittance trend map. Experimental results show that the transmittance trend map has a better effect on fog removal and improves the problem of dark brightness and over-saturation of the original algorithm. (3) aiming at the problem of low efficiency of the original algorithm, this paper proposes two improved methods of filtering smoothing based on the concept of transmittance trend map: first, based on the concept of Gao Si smoothing filter, Based on the improved Gao Si spatial smoothing filter, the transmittance trend map is obtained. Secondly, based on the improved guided filtering method, by reconstructing the guide matrix to smooth the coarse transmittance map, not only the equivalent and delicate transmittance trend map can be obtained, but also the time cost of the traditional guided filter to optimize the transmittance can be reduced. The efficiency of the algorithm is 3 times that of the original algorithm. (4) based on the evaluation method of visible edge enhancement, by introducing the restoration index of color information and combining the definition index, the comprehensive evaluation index of image de-fogging quality without reference is constructed to evaluate the result of de-fogging more comprehensively. Based on the Matlab platform, this paper compiles the related algorithm program, and simulates the above algorithm. The experimental results show that the proposed scheme can effectively improve and solve the problems of the traditional dark color priori de-fogging algorithm.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】