Dark Channel Prior And Haze-Line Prior Based Real-time And Adaptive Video Dehazing Algorithm

There are plenty of outdoor video systems in our daily life,such as vehicle supervisory systems and vehicle self-driving systems.Under foggy or hazy conditions,the visibility of videos and the performance of pattern recognition algorithms will reduce,which makes these outdoor video systems to be less usable.In order to improve the usability of these systems under hazy condition,we propose a complete and robust solution which can adaptively dehaze videos in real time.We first propose a real-time image dehazing algorithm using both dark channel prior and Haze-Line prior.This algorithm uses dark channel prior to estimate local transmission and uses Haze-Line prior to estimate non-local transmission.We find out that local transmission and non-local transmission have a complementary relationship.Moreover,we propose to use the superposition of local transmission and non-local transmission to replace the smoothing process,which not only can significantly improve the running speed,but also can overcome the drawbacks of other dehazing algorithms that only use dark channel prior or Haze-Line prior.Experimental results show that the proposed dehazing algorithm has a high running speed and an ideal dehazing effect.We also propose an adaptive dehazing framework which can be used along with any dehazing algorithm based on hazy image formation model.Via the cooperation of this framework and the native dehazing algorithm,we can dehaze images adaptively.This framework uses information loss to decide whether estimated transmission is proper,and sends the feedback to the native dehazing algorithm.The native dehazing algorithm then uses this feedback to adaptively adjust its parameters.When using this framework on videos,this framework hardly costs any extra running time by using a proposed acceleration method.Experimental results show that the cooperation of proposed adaptive dehazing framework and proposed dehazing algorithm can yield an ideal adaptive dehazing effect.At last,we propose a reliable evaluation method.This evaluation method uses stereo datasets and some widely-used classic equations to generate reliable dehazing datasets,and then uses the error between dehazing results and corresponding ground truth to evaluate dehazing algorithms reliably.