基于统计量的图像去雾算法研究

发布时间：2018-11-21 15:35

【摘要】：图像处理旨在突出图像中的部分细节信息,以便于人眼的视觉观察和计算机后续分析操作.在雾霾环境条件下,由于大气粒子的作用,户外图像的能见度受到限制且图像的对比度有所下降.为了解决雾天图像质量降低的问题,目前大多数处理方案均采取基于图像增强和图像复原的去雾算法,图像增强算法通过提高图像的对比度以达到清晰化效果;图像复原以有雾图像成像模型为基础,对大气散射机制建模,恢复出无雾的可能清晰图像.本文基于雾天图像形成模型和退化机理,探索了图像去雾的关键技术和实现方法,并给出了暗通道先验的统计证据和非局部去雾的改进算法.论文的主要工作如下:(1)暗通道先验是一种基于室外清晰无雾图像的统计规律,即大部分户外无雾图像的非天空局部区域都存在一些至少一个颜色通道强度很低的像素.本文从统计的角度出发,假定三通道相互独立、场景点与其所在领域内其它像素点相互独立,将彩色无雾图像均一化后的RGB值看作统计变量.假定这三个变量均服从贝塔分布,给出了进行两次最小值滤波(先对RGB三通道取小,再在一个邻域内取小)后变量的密度函数和分布函数,从而验证暗通道先验的有效性.(2)现有的图像去雾方法依据所采用先验信息的不同而大体分为局部和非局部两类.Berman等人基于清晰图像在RGB空间的非局部聚类特性,构造了有雾图像每一颜色类的几何表示一雾线(Haze-Line),雾线的最大辐射坐标(LRC:Largest Radial Coordinate)是估计初始透射率的关键.本文从统计学的观点出发,给出了 LRC的一个无偏估计.实验结果表明,该方法可得到至少与原方法相当的结果.
[Abstract]:Image processing aims to highlight some details in the image for visual observation and computer analysis. Under the condition of haze, the visibility of outdoor images is restricted and the contrast of the images is decreased due to the action of atmospheric particles. In order to solve the problem of image quality degradation, most of the current processing schemes are based on image enhancement and image restoration de-fog algorithm, image enhancement algorithm by improving the contrast of the image to achieve clarity effect; Based on the foggy image imaging model, the atmospheric scattering mechanism is modeled to restore the possible clear image without fog. Based on the fog image formation model and degradation mechanism, this paper explores the key techniques and implementation methods of image de-fogging, and gives the prior statistical evidence of dark channel and the improved non-local de-fogging algorithm. The main work of this paper is as follows: (1) the dark channel priori is a statistical rule based on the clear outdoor fog-free images, that is, there are at least one pixel with low intensity of color channel in the non-sky region of most outdoor fog-free images. In this paper, we assume that the three channels are independent of each other, and the scene points are independent with other pixel points in the field. The RGB value of the color fog free image is considered as a statistical variable. Assuming that these three variables are all subject to the Beta distribution, the density function and distribution function of the variables are given after the minimum filtering of two times (the RGB three-channel is small, then the RGB is small in a neighborhood). In order to verify the validity of dark channel priori. (2) the existing image de-fogging methods can be divided into local and non-local categories according to the different prior information used. Berman et al., based on the non-local clustering characteristics of clear images in RGB space, The geometric representation of a fog line (Haze-Line) for each color class of a fog image is constructed. The maximum radiative coordinates of the fog line (LRC:Largest Radial Coordinate) are the key to estimate the initial transmittance. In this paper, an unbiased estimate of LRC is given from the point of view of statistics. The experimental results show that the proposed method can obtain at least the same results as the original method.
【学位授予单位】：宁夏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】