首页 > 硕士 > 军事学 > 正文

基于滤波的暗原色先验图像云雾算法

Image Dehazing Algorithms Using Dark Channel Prior Based on Filtering

作者: 专业:密码学 导师:和红杰 年度:2015 学位:硕士  院校: 西南交通大学

Keywords

image dehazing, dark channel prior, transmission, halo effects, color fidelity, filter

        众所周知,在含雾天气下,人们基本无法看清前方的物体,容易造成各种交通事故,给人们的日常生活带来诸多不便,且由于雾天图像的低对比度特性使得研究者也无法较好的利用这类图像获得有用的信息。图像去雾算法将一幅带雾图像变成清晰无雾图像,广泛的应用于研究领域的各个方面,如目标识别、目标检测等。本文主要研究了基于暗原色先验的单幅图像快速去雾算法,从以下几个方面进行阐述:1)指出现有基于暗原色先验算法得到的去雾结果存在光晕现象及颜色失真等问题,提出一种基于边界邻域最大值滤波的图像快速去雾算法。首先,通过边缘检测寻找图像边界被低估的暗原色值并对其进行边界邻域最大值滤波,以得到更为准确的透射率图来消除光晕现象;其次,对暗原色图乘以一个尺度因子,扩大透射率的取值范围,增强去雾结果的对比度;最后,通过设置两个亮度阈值以及一个平坦阈值,排除图像中高亮度物体的影响,获得更为准确的大气光值,使得去雾结果颜色保真度较高。仿真结果表明,本文算法不仅可以去除光晕现象,获得较好的去雾图像,同时也大大的缩减了时间复杂度。2)针对基于边界邻域最大值滤波的图像去雾算法不能自适应求取透射率及去雾结果往往偏黑,设计一种自适应求取透射率及自适应亮度拉伸的图像去雾算法。该算法通过计算已知暗原色点结合透射率方程获取大气散射系数,估计透射率。针对去雾图像直方图的分布,决定是否对去雾图像进行亮度拉伸。实验结果表明,该算法不仅可以消除光晕现象,获得高亮度及高颜色保真度的去雾结果,同时针对公路图像以及遥感图像的去雾也获得了良好的效果。3)由于上述方法不能很好的处理深度不连续区域,尤其是细小缝隙区域的雾。因此,本文使用快速加权中值滤波进行优化透射率,该算法充分利用了快速加权中值滤波器的优点,即保持边缘与角点以及速度快。实验结果表明,该方法对含细小缝隙的图像,去雾效果较理想。除此之外,去雾速度也进一步提高。
    As we all know, people can not see clearly in the foggy day. As a result, lots of traffic accidents will be happened due to the fog and it brings inconvenience to people’s daily life. For researchers, owning to the low contrast and blur pictures taken in the foggy day, they can not get the useful information from the haze image. Image dehazing algorithm is to make a haze image becomes more clear. It has been widely used in many areas, such as target recognition, object detection, etc. This paper mainly studies the fast haze removal from single image based on dark channel prior and it will be discussed from the following aspects:1) Pointed out the presence of halo effects and color distortion problem in the existing algorithm based on the dark channel prior. Thereby a fast single image defogging algorithm based on edge-maximum filter has been proposed. Firstly, an edge-maximum filter is used to recover the undervalued dark pixels obtained by edge detection, which is to receive an accurate transmission map and eliminate the halo effects. Then in order to gain a high contrast dehazing image, all the dark pixels are multiplied by a scaling factor to improve the dynamic ranges of the transmission. Finally, two brightness thresholds and one flat threshold are set to eliminate the influence of high light objects in the image and obtain a more accurate airlight, which keeps a high color fidelity in the dehazing image. The simulation results show that the proposed method, compared with other algorithms, could eliminate the halo effects and achieve the dehazing image with high contrast and high color fidelity, especially for the images containing high light objects or rich details. Meanwhile, the computational speed is also improved.2) The dehazing algorithm which used the edge-maximum filter can not acquire the transmission adaptively and the dehazing image always looks dark. Consequently, a self-adaptive transmission estimation and brightness stretching for image dehazing algorithm has been came up with to solve the problem. It combined the known dark channel points with transmission equation to obtained the atmospheric scattering coefficient and then calculated the accurate transmission. Finanly, through the histogram distribution of dehazing image, it decide whether to stretch the dehazing image’s brightness. The simulation results show that the proposed method could eliminate the halo effects and achieve the dehazing image with high contrast and high color fidelity. Meanwhile, it can also dehazing some specific images such as road images and remote sensing images.3) Above methods can not handle the fog image which have depth-discontinuities, especially some small gaps. To address this problem, we using the fast weighted median filter to optimize the transmission. This algorithm makes full use of the advantages of rapid weighted median filter, which maintains the edge and corner, and fast. Experimental results show that, with this method, the dehazing result will be better in the fog image which includes some small gaps, and the speed is further improved.
        

基于滤波的暗原色先验图像云雾算法

摘要6-7
英文摘要7-8
第1章 绪论11-17
    1.1 课题的研究背景及意义11-12
    1.2 图像去雾的国内外研究现状12-14
        1.2.1 基于非物理模型的单幅图像去雾12-13
        1.2.2 基于物理模型的单幅图像去雾13-14
    1.3 去雾结果评价方法14-15
    1.4 本文组织结构15-17
第2章 暗原色先验的图像去雾17-29
    2.1 图像去雾的基本原理与模型17-18
    2.2 暗原色先验图像去雾算法介绍18-23
        2.2.1 暗原色先验18-19
        2.2.2 透射率的估计19-20
        2.2.3 透射率的优化20-22
        2.2.4 图像去雾22
        2.2.5 存在的问题及原因22-23
    2.3 现有基于暗原色先验的图像去雾改进算法介绍23-28
        2.3.1 针对透射率的改进24-25
        2.3.2 针对大气光的改进25-26
        2.3.3 去雾结果后处理26-28
        2.3.4 现有算法存在的不足28
    2.4 本章小结28-29
第3章 改进的暗原色先验图像快速去雾29-44
    3.1 引言29
    3.2 基于边界邻域最大值滤波的图像快速去雾算法29-35
        3.2.1 大气光的改进29-31
        3.2.2 透射率的改进31-33
        3.2.3 实验结果与分析33-35
    3.3 基于自适应透射率估计及亮度拉伸的图像去雾35-42
        3.3.1 自适应透射率求取35-37
        3.3.2 自适应亮度拉伸37
        3.3.3 实验结果与分析37-42
    3.4 本章小结42-44
第4章 基于边缘角点滤波器的图像去雾算法44-55
    4.1 引言44
    4.2 基于快速加权中值滤波器的图像去雾算法44-48
        4.2.1 快速加权中值滤波器的介绍44-46
        4.2.2 基于快速加权中值滤波器的透射率估计46-48
    4.3 实验结果与分析48-54
    4.4 本章小结54-55
第5章 单幅图像快速去雾算法的可视化演示55-60
    5.1 MATLAB GUI介绍55-56
    5.2 基于暗原色先验的单幅图像快速去雾软件56-60
        5.2.1 输入图像56-57
        5.2.2 单幅图像快速去雾57-58
        5.2.3 计算客观评价指标58-59
        5.2.4 保存去雾结果及退出软件59-60
结论与展望60-62
致谢62-63
参考文献63-68
攻读硕士学位期间发表的论文及参与的科研项目68
        下载全文需50


本文地址:

上一篇:安全云存储的应用研究
下一篇:最后一页

分享到: 分享基于滤波的暗原色先验图像云雾算法到腾讯微博           收藏
评论排行
公告