雾霾图像增强处理技术研究

发布时间：2018-05-30 10:13

本文选题：雾霾图像 + 图像增强　；参考：《哈尔滨理工大学》2017年硕士论文

【摘要】：近几年雾霾天在我国北方出现次数极为频繁,尤其是在冬季供暖时期,雾霾情况尤为严重。雾霾条件下由于空气污染物导致能见度剧烈下降,使得室外图像采集系统采集得到的图像退化模糊,清晰度下降,很多图像的重要特征信息丢失,给图像信息提取工作带来极大困难,同时也对室外物体追踪,交通检查等工作造成限制与影响,因此开展雾霾天的降质图像增强技术研究具有十分重要的研究价值和实际应用意义。近些年来数字图像处理技术飞速发展,处理雾霾图像的方法也衍生出很多。大致上分为两类:一类是通过数字图像处理的方法进行处理,退化的图像一般亮度较暗,对比度较低,图像增强方法不考虑图像退化原因,从图像亮度和对比度角度对图像进行增强;另一类是通过了解雾霾图像质量退化的原因,建立数学模型,从而实现对图像的去雾处理。传统的图像增强方法在增强图像的同时也丢失了图像的细节、增强了图像中包含的噪声信息,导致信息熵下降。本文从雾天图像增强角度出发,以此为研究主线,探讨研究雾霾图像增强算法的基本框架,并对算法中的关键技术展开深入的研究学习,对传统图像增强算法存在的缺点进行改进。在基于Retinex算法的图像增强方法中,在传统的多尺度Retinex算法(Multi-Scale Retinex,MSR)的基础上,改进了原有的基本算法,进行实验仿真得到的处理结果和算法速率,然后对比分析改进后算法与已有的算法。在基于小波变换的图像增强方法中,结合小波变换多尺度、多分辨率的特点与对比度受限自适应直方图均衡的特性,提出一种离散小波变换与对比度受限自适应直方图均衡相结合的图像增强算法。实验结果表明,这两种方法能有效增强图像,相对来说失真比较小,实现雾霾天降质图像的增强。
[Abstract]:In recent years, haze days appear frequently in the north of China, especially in winter heating period, the haze situation is especially serious. Under the condition of haze, the visibility of the image acquired by the outdoor image acquisition system is degraded and blurred due to air pollutants, resulting in the loss of many important features of the image. It brings great difficulties to the extraction of image information, at the same time, it also limits and affects the work of outdoor object tracking, traffic inspection and so on. Therefore, it is of great value and practical significance to carry out the research of degraded image enhancement in haze days. With the rapid development of digital image processing technology in recent years, there are many methods for processing haze images. It can be divided into two categories: one is processed by digital image processing, the degraded image is generally dark brightness, low contrast, image enhancement method does not consider the reason of image degradation. The image is enhanced from the perspective of image brightness and contrast, and the other is to realize the image de-fog processing by understanding the causes of the degradation of haze image quality and establishing a mathematical model. Traditional image enhancement methods not only enhance the image, but also lose the details of the image, enhance the noise information contained in the image, resulting in the decrease of information entropy. From the point of view of fog image enhancement, this paper discusses the basic framework of haze image enhancement algorithm, and studies the key technology of the algorithm. The shortcomings of traditional image enhancement algorithms are improved. In the image enhancement method based on Retinex algorithm, based on the traditional multi-scale Retinex algorithm (Multi-Scale Retinexus MSRs), the original basic algorithm is improved, and the processing results and algorithm speed obtained by experimental simulation are obtained. Then the improved algorithm and the existing algorithm are compared and analyzed. In the image enhancement method based on wavelet transform, combining the characteristics of multi-scale and multi-resolution of wavelet transform and the characteristic of contrast constrained adaptive histogram equalization, An image enhancement algorithm based on discrete wavelet transform and contrast constrained adaptive histogram equalization is proposed. The experimental results show that the two methods can effectively enhance the image, and the distortion is relatively small, so the enhancement of the haze image can be realized.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】