Research On Real-time Reatoration Of Atomized And Degraded Optical Image

In hazy weather,the scattering of light by the suspended particles in the air makes the captured image with low contrast,which reduces the visibility and clarity of the image,causing the loss of image detail information.At the same time,the ambiguity of the input image will directly affect the extraction and processing of subsequent information.In order to obtain clear and natural video images,real-time image dehazing algorithm has become a research hotspot.The current research mainly focuses on the simple scene with background static and foreground changing dynamically.However,as for the complex scenes with both dynamic changes in foreground and background,real-time image dehazing algorithms are less researched and difficult.To this end,the purpose of this project is to explore a technical approach that can achieve real-time haze removal of complex scene video images.To this end,a new method of real-time fog elimination in complex scene video images has been explored.The main research contents of this dissertation include:First of all,the atmospheric scattering model is introduced.After that,the propagation characteristics of haze daylight and the reason of the degradation of atomized image are analyzed.Based on this,we study the dark channel prior dehazing algorithm and implement the algorithm based on CPU platform and C ++ language.Secondly,aiming at the problem that the existing image de-fogging algorithm is time-consuming and difficult to meet the real-time requirement on the CPU platform,this paper uses the high-performance GPU platform and CUDA programming language to speedup parallelization of dark channel prior dehazing algorithm,and optimize it by sharing memory,register and multiple CUDA streams.The experimental results show that the performance efficiency of the image dehazing algorithm is greatly improved after the parallel acceleration of the dark channel prior fog algorithm based on the GPU platform and the CUDA language.Finally,the real-time dehazing technique for video images of scenes the dynamic change of foreground and background studied.For full HD video images,a dual GPU acceleration method is proposed.In order to seek the optimal control mode of parallel acceleration of dual-GPU,three strategies of sub-frame,sub-step and sub-pixel are proposed.The sub-frame is to speed up the adjacent frame video images with one GPU respectively.The sub-step is to divide the single frame image defogging algorithm into two parts and use one GPU to accelerate separately.Sub-pixel is to divide the haze image data into two parts and use one GPU to accelerate apart.The experimental results show that among the three implementation strategies of parallel execution of dual GPU multi-threading,the framing method has the fastest dehazing speed.As for the frame resolution of 1920x1080,the dehazing speed is over 25 frames per second,which can meet the real-time dehazing requirements.