导　　师： 庞彦伟

学科专业： 0810

授予学位： 硕士

作　　者： ;

机构地区： 天津大学

摘　　要： 在雾或雾霾天气条件下，受大气散射效应的影响，户外场景的能见度很低。这不仅给人们的生活和工作带来了不便，还使得拍摄的图像严重退化。雾天图像模糊不清、颜色失真、对比度低，图像中的细节信息无法清晰体现出来。这些都大大降低了雾天图像的利用价值，干扰了很多领域的正常工作。再加上我国近几年雾霾天气频发，图像去雾算法的研究已经成为计算机视觉和图像处理领域的热点研究问题之一，具有十分深远的实际意义。高效的图像去雾算法可以应用在监控系统和模式识别等领域。 本文总结了图像去雾的国内外研究现状，然后对传统基于暗原色先验的去雾算法及其改进算法进行了详细的介绍。传统基于暗原色先验的去雾算法利用了雾天图像成像模型，可以取得非常好的去雾效果；它的改进算法采用快速双线性滤波加速整个去雾过程，在去雾效率上得到了很大的提升。本文实现了这两种算法，并将快速双线性滤波移植到GPU上获得进一步的加速，之后又进行了大量的实验，分析了这两种算法的优点和不足。 基于暗原色先验的算法利用了物理模型，是当前最好的去雾算法。尽管如此，本文提出了一种非模型算法，叫做iItem（intensity Inverting and TonE Mapping，强度反转和色调映射）。受Dong等人的工作/[1/]启发，本文发现，反转强度后的有雾图像与低照度图像之间存在着相似性。iItem算法以这个观察为基础，利用色调映射增强反转强度后的有雾图像，之后再次执行反转得到去雾结果。然而，单纯的强度反转和色调映射并不能得到令人满意的去雾效果。因此，本文在色调映射之前加入了基于最大值的分块和双线性滤波，并修改了面向低照度图像的色调映射方法，从而成功实现了图像去雾。实验结果表明，在主、客观评价指标上，本文提出的iItem算法比基于暗原色先验的算法都具有优势。 In fog or haze weather conditions, the visibility of outdoor scene is poor, due tothe atmospheric scattering effect. This not only brings inconvenience to people’s lifeand work, but also makes the captured image degraded severely. With low contrast,distorted color and hidden details, the hazy image is blurred, which greatly reduces itsvalue and jams many fields’ work. Moreover, haze weather in our country is frequentrecently. Study on dehazing has become a hot topic in the field of computer visionand image processing, with profound practical significance. Effective dehazingalgorithm can be applied in areas such as monitoring system and pattern recognition. This paper summarizes the research status of haze removal algorithm, followed adescription of traditional haze removal algorithm using dark channel prior and itsimprovement. The former algorithm adopts physical model and works very well,while the latter uses fast bilateral filtering to accelerate dehazing process and achievesgreat improvement. This paper implements these algorithms and performs fastbilateral filtering on GPU for further acceleration. Then many experiments are carriedout and the strengths and weaknesses of these algorithms are analyzed. The dehazing algorithm based on dark channel prior adopts physical model and itis current state-of-the-art dehazing algorithm. Nevertheless, inspired by Dong et al.’swork/[1/], this paper proposes a model-free one called iItem /(intensity Inverting andTonE Mapping/), which is based on the observation that intensity-inverted hazy imagehas similarity with low-light image. It implements tone mapping on intensity-invertedhazy image followed by intensity re-inverting to obtain dehazed result. However,simple intensity inversion plus tone mapping does not bring satisfying result.Therefore, this paper adds maximum-based blocking and bilateral filtering before tonemapping and modifies this low-light oriented tone mapping to achieve successfuldehazing. Experimental results show that iItem is comparable with and even betterthan the algorithm based on dark channel prior in subjective and objective evaluation.

关 键 词： 图像去雾 色调映射 图像增强 双线性滤波

分 类 号： [TP391.41]

领　　域： [自动化与计算机技术] [自动化与计算机技术]