基于暗通道先验的图像去雾改进算法研究

发布时间：2019-06-24 20:37

【摘要】：近些年来,随着城市中车辆和人口密度不多增加,空气质量遭到严重破坏,导致雾霾现象频发。雾霾的存在,致使能见度大大降低,图像采集设备无法直接获得准确的信息,进而对周围环境作出错误的判断,严重的甚至会导致灾难发生。雾天情况下,由于空气中漂浮着大量悬浮颗粒,光线在传播过程中会被散射而造成强度衰减,光线强度的衰减使得图像对比度下降,细节模糊,色彩保真度下降。因此,雾霾图像的清晰化处理具有重要的现实意义,图像去雾领域越来越受到国内外研究者的关注。在众多研究成果中,何恺明博士的暗通道去雾算法给图像去雾技术领域带来了新的启发,该算法具有简单有效、实时性和自动去雾的优点,其缺点主要是对大气散射物理模型过度依赖、无法选取相匹配的滤波模板大小、对天空缺乏适用性以及恢复后图像偏暗。本文在大气散射物理模型基础上,分析了雾天图像质量下降的原因,并针对暗通道去雾算法进行实验、改进与尝试,并取得了较好的成像效果。本论文的工作主要体现在以下几点:(1)提出了一种基于多尺度的思想来细化透射率的方法,从而达到去雾效果。该方法以暗通道先验准则为理论基础,在保证暗通道先验准则与区域内的透射率相同假设成立的同时,防止估计的透射率图像在暗通道突变的区域发生块状效应。(2)在求取暗通道时采用自适应的方法进行滤波,在提升对比度的同时保持了结构信息。对原始的大气光值求取过程进行改进,有效抑制了天空区域去雾后色彩过饱和的现象。(3)对于透射率的估计,为了克服天空区域复原之后呈现彩色的光晕问题,本文通过分析透射率特点,在获得初始透射率之前的中间过程进行特殊处理,然后对其进行引导滤波优化,最终获得了较好的复原效果。实验结果充分表明了本文提出的改进方法良好的改进效果,且较其他改进方法具有简单并且更有效的特点,但是本文的改进方法并不能适用各类含雾图像,尤其是对含有大量天空或明亮区域的图像,去雾效果并不理想,值得继续深入研究和改进。
[Abstract]:In recent years, with the small increase of vehicle and population density in the city, the air quality has been seriously damaged, resulting in the frequent occurrence of haze phenomenon. Because of the existence of haze, the visibility is greatly reduced, and the image acquisition equipment can not directly obtain accurate information, and then make a wrong judgment on the surrounding environment, which will even lead to disaster. In the case of fog, due to the floating of a large number of suspended particles in the air, light will be scattered in the process of propagation, resulting in intensity attenuation, the attenuation of light intensity makes the image contrast decrease, the details are blurred, and the color fidelity decreases. Therefore, the clarity of haze image processing is of great practical significance, and the field of image fog removal has been paid more and more attention by researchers at home and abroad. Among the many research results, Dr. he Kaiming's dark channel fog removal algorithm has brought new inspiration to the field of image defogging technology. the algorithm has the advantages of simple and effective, real-time and automatic fog removal. The main disadvantage of the algorithm is that it is too dependent on the physical model of atmospheric scattering, unable to select the matching filter template size, lack of applicability to the sky and dark image after restoration. In this paper, based on the physical model of atmospheric scattering, the reasons for the decline of image quality in fog days are analyzed, and the dark channel fog removal algorithm is tested, improved and tried, and good imaging results are obtained. The work of this paper is mainly reflected in the following points: (1) A method based on multi-scale idea to refine the transmittance is proposed to achieve the effect of fog removal. Based on the theory of dark channel prior criterion, while ensuring that the dark channel prior criterion is the same as the transmittance assumption in the region, the estimated transmittance image is prevented from block effect in the region of dark channel mutation. (2) the adaptive method is used to filter the dark channel, which not only improves the contrast, but also maintains the structure information. The original atmospheric light value calculation process is improved, which effectively suppresses the phenomenon of color supersaturation after fog removal in the sky region. (3) for the estimation of transmittance, in order to overcome the color halo problem after sky region restoration, this paper analyzes the transmittance characteristics, carries on the special treatment to the intermediate process before obtaining the initial transmittance, and then carries on the guide filter optimization to it, and finally obtains the better restoration effect. The experimental results fully show that the improved method proposed in this paper has a good improvement effect, and has the characteristics of simple and more effective than other improved methods, but the improved method in this paper is not suitable for all kinds of fog images, especially for images with a large number of sky or bright areas, and the fog removal effect is not ideal, which is worthy of further study and improvement.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】