基于暗通道先验的去雾技术研究

发布时间：2018-04-09 20:32

本文选题：图像去雾　切入点：暗通道先验　出处：《中国科学院大学(中国科学院光电技术研究所)》2017年硕士论文

【摘要】：现今环境问题日益严峻,雾霾天气频繁出现,不仅给人类的生存坏境带来威胁,同时影响到社会的方方面面,用成像设备拍摄,会发现获取的图像明显的退化,包括颜色出现失真,对比度有所降低,细节也变得模糊不清等,会对后续的处理工作造成严重的影响,给信息采集,目标识别提取等带来了一系列的困难因此,对图像去雾技术进行研究十分必要。本文主要从以下几个方面开展研究:(1)对国内外图像去雾技术进行广泛调研,将去雾方法分为两大类:基于增强和基于复原的去雾,基于增强的去雾方法包括直方图均衡化,小波变换等,基于复原的包括利用偏微分方程,利用景深估计和利用先验信息等。通过对各类算法进行仿真,对其优缺点进行详细的分析,发现暗通道先验去雾算法对大多数自然场景图像有着较好的效果,所以选取暗通道先验的去雾算法作为后续的重点研究。(2)对一些基础理论进行研究,包括雾的定义,雾气形成所需的条件,雾天成像模型,雾天图像特性,暗通道先验原理,去雾评价指标(主客观)。(3)研究暗通道先验去雾算法,对算法流程进行了详细的分析,包括求取暗通道,估计大气光,估计透射率,优化透射率。在对暗通道先验去雾算法仿真过程中,发现了每一步的不足之处,进而针对存在的问题与缺陷,对每一步引入相应的改进算法,在视觉效果和时间复杂度两个方面得到了相应的改善。(4)针对暗通道先验去雾算法效率低的问题,提出基于暗通道先验的HSI空间快速去雾算法。首先将有雾图像从RGB空间转换到HSI空间并进行分块,对每一区块求取暗通道点,然后对暗通道点进行饱和度矫正,再由HSI空间转换到RGB空间,求得图像中每个暗通道点的雾气分量,通过插值的方法,得到全局雾气分量图,最后由有雾图像减去全局雾气分量图,并进行亮度调整,最终得到复原后的图像。(5)本文图像仿真均在MATLAB平台上实现,分别对暗通道先验去雾算法及针对每一步的改进算法进行了仿真,且对提出的基于暗通道的快速去雾算法进行了仿真,通过评价指标可以看出,以上改进算法均得到了较好的预期效果。
[Abstract]:Nowadays, the environmental problems are becoming more and more serious, and the haze weather frequently appears, which not only threatens the survival of human beings, but also affects all aspects of society.Including color distortion, lower contrast, blurred details and so on, which will have a serious impact on the subsequent processing work and bring a series of difficulties to information collection, target recognition and extraction, etc.It is necessary to study the image de-fogging technology.In this paper, we mainly carry out research on image de-fogging technology in China and abroad from the following several aspects. We divide the de-fogging methods into two categories: based on enhancement and restoration, and based on enhancement, which includes histogram equalization.Wavelet transform is based on partial differential equation, depth of field estimation and prior information.Through the simulation of all kinds of algorithms and the detailed analysis of their advantages and disadvantages, it is found that the dark channel priori de-fogging algorithm has a good effect on most natural scene images.So select the dark channel priori de-fogging algorithm as the focus of the following research.) to study some basic theories, including the definition of fog, the conditions required for fog formation, fog imaging model, fog image characteristics, dark channel prior principle,A priori de-fogging algorithm for dark channels is studied in this paper. The flow of the algorithm is analyzed in detail, including the selection of dark channels, estimation of atmospheric light, estimation of transmittance, and optimization of transmittance.In the simulation of dark channel priori de-fogging algorithm, the shortcomings of each step are found, and then the corresponding improved algorithm is introduced for each step in view of the existing problems and defects.The visual effect and time complexity are improved accordingly. Aiming at the low efficiency of dark channel priori de-fogging algorithm, a fast de-fogging algorithm in HSI space based on dark channel priori is proposed.Firstly, the foggy image is transformed from RGB space to HSI space and divided into blocks. The dark channel points are obtained for each block, then the saturation of dark channel points is corrected, and then converted from HSI space to RGB space.The fog component of each dark channel point in the image is obtained. By interpolation, the global fog component map is obtained. Finally, the global fog component image is subtracted from the fog image, and the brightness is adjusted.Finally, the reconstructed image. 5) the image simulation in this paper is implemented on the MATLAB platform. The prior de-fogging algorithm of dark channel and the improved algorithm for each step are simulated, respectively.The proposed fast de-fogging algorithm based on dark channel is simulated, and it can be seen from the evaluation index that the improved algorithm has better expected results.
【学位授予单位】：中国科学院大学(中国科学院光电技术研究所)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】