视频图像去雨技术研究前沿  

The research forefront of rain removal for videos and images

作　　者：徐波[1,2] 朱青松[2] 熊艳海[2,3] 

机构地区：[1]中国科学技术大学计算机科学与技术学院,合肥230027 [2]中国科学院深圳先进技术研究院,广东深圳518055 [3]新加坡南洋理工大学交叉学科学院,新加坡639798

出　　处：《中国科技论文》2015年第8期916-927,共12页China Sciencepaper

基　　金：国家重点基础研究发展计划(973计划)资助项目(2010CB732606);国家自然科学基金资助项目(61303166)

摘　　要：户外视觉系统越来越广泛地应用于军事、交通及安全监控等领域,但是恶劣天气严重影响了系统的性能,而下雨是最频繁的恶劣天气之一,会严重损害视频图像的质量,因此检测和去除视频图像中的雨滴对于一个全天候的户外视觉系统来说是必不可少的。去雨技术不仅可以恢复被雨滴影响的视频图像,而且有利于对视频图像的进一步处理,包括基于视频图像的目标检测、识别、追踪、分割和监控等技术的性能提高。为了去除视频图像中的雨滴,首先要分析其成像规律,深入研究雨滴的几何、亮度、色彩和时空等特性,然后基于这些特性进行雨滴检测,最后去除雨滴,修复视频图像。讨论了目前的各种去雨技术,从视频图像中雨滴的特性入手,详细地阐述了各类去雨算法及其优缺点。为了内容的全面性和完整性,对部分具有代表性的算法进行了定量和定性分析。最后对该研究领域中亟待解决的问题进行了总结,并且在此基础上对其未来的发展方向做了进一步展望。Outdoor vision systems are becoming increasingly widely used in the field of military, transportation, security surveil- lance applications. However, bad weathers affect severely the performance of system. Rain is also the most frequent adverse weather, and the video image influenced by the rainfall will be severely degraded. Therefore, it is essential for an all-weather out- door vision system to detect and remove raindrops in the videos and images. Rain removal technique not only restores the rain-af- fected videos and images, but also is beneficial to the further processing of videos and images, including increasing performance of computer vision algorithms in areas such as object detection, recognition, tracking, segmentation and video surveillance. In order to restore rain affected pixels in images, the first step is to analyze the process of rain projection on image and the geometric, in- tensity, chromatic and spatiotemporal properties of raindrop. These properties are applied to detect and remove the raindrop to restore image sequenee. We made an overall introduction on state-of-art rain removal techniques, a detailed presentation of differ- ent kinds of algorithms as well as their merits and demerits. For comprehensiveness and completeness of the contents, some rep- resentative algorithms were analyzed quantitatively and qualitatively. Finally, we make a conclusion on the urgent problems on this topic, and then we propose an assumption on potential rseearch trends related to it.

关 键 词：计算机视觉 图像修复 雨滴检测 雨滴去除 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]