弱光低照多光源发光互扰尘雾图像的退化模型及清晰化方法研究

A clear visible image is very important for traffic navigation,security monitoring in haze/dust weather at night, and safety monitoring for various construction underground.However, all existing machine vision systems face the following common image degradation problems in above mentioned practical applications:①poor illumination and poor contrast caused by weaken light;②Over-high-brightness caused by mutual interference from multi-source of light;③low contrast caused by dust and water vapor particles in the air. In this project, the key theories and methods involved in solving the above problems are studied. Firstly, the image degradation model under low light and low illumination, multi-source luminous interference and dust-fog environment is studied, which provides the model basis for solving the image degradation problem under this complex condition. Secondly, aiming at the over high brightness caused by the multi-source luminous interaction, an image enhancement method for over- high-bright is proposed in an effort to enhance the visibility of over-high-bright scene area; third, a method for enhancing image brightness and contrast is studied in order to improve the overall visibility of low-brightness image. Finally,we evaluate the proposed methods from qualitative and quantitative perspective. This project will provide new theories and methods for solving the problem of image blur in complex environment. It is also of great significance to improve the application ability of machine vision system in complex environment.

清晰的图像画面对提高机器视觉在实际中的应用性能非常关键。然而在夜间有霾或沙尘天气下的交通导航、安防监控，及城市地下交通建设、铁路隧洞挖掘、矿产开采等地下施工的安全监控应用中，现有的机器视觉系统均面临如下图像问题的共同挑战：①弱光低照引起的低亮度、低对比度；②多光源发光互扰引起的高亮光模糊；③灰尘及水汽等悬浮颗粒引起的尘雾模糊。本项目针对上述问题求解所涉及关键理论及方法进行研究。首先，研究弱光低照、多光源发光互扰及尘雾环境下的图像退化模型，为研究此类复杂条件下图像退化问题的解决方法提供模型依据；其次，研究场景能见度的图像增强方法，解决弱光低照及多光源发光互扰引起的能见度低的问题，最后从主客观两方面对上述研究进行分析评价。本研究将为解决复杂环境下的图像退化问题提供新的理论和方法，对提高复杂环境下机器视觉的应用能力具有重要意义。

清晰的图像画面对机器视觉系统在实际中的应用至关重要。然而无论是在夜间有霾或沙尘天气下的交通导航、安防监控，还是在城市地下交通建设、或铁路隧洞挖掘、或矿产开采等各种地下施工的安全监控应用中，现有的机器视觉系统都面临如下几个图像问题的共同挑战：①弱光低照引起的低亮度、低对比度；②多光源发光互扰引起的高亮光模糊；③灰尘及水汽等悬浮颗粒引起的尘雾模糊。本项目针对上述问题求解所涉及关键理论及方法进行研究。主要研究内容和成果如下：1）分别构建了低照环境下雨雾图像退化模型和沙尘雾霾图像退化模型；2）提出基于深度集成网络的低照雨雾图像增强方法；3）提出基于halo抑制的DCP沙尘图像增强方法；4））提出一种带颜色恢复的沙尘图像限制性直方图均衡增强方法；5）提出一种基于直方图相似先验及边界约束的沙尘图像增强方法；（6）提出一种基于上下文引导的生成对抗网络图像去雾霾方法。相关研究内容在国内外学术期刊、会议上发表及投稿学术研究论文5篇，其中SCI检索3篇，培养博士生1名、硕士生5名。.本项目研究成果将为解决弱光低照与多光源互扰及尘雾环境下的图像清晰化复原问题提供新的理论和方法，对提高计算机视觉系统在复杂环境下的应用具有重要意义。该研究成果与我国的经济和社会发展密切相关，对交通导航、安防监控、遥感监测等技术领域都有着积极的指导作用。