Image Dehazing Algorithm Research And Its Application

Outdoor images obtained in adverse weather(e.g.,foggy or hazy)usually lose contrast and fidelity.Most automatic systems for surveillance,navigation,aviation and intelligent vehicles,etc.,work well when the input images have clear visibility.Therefore,improving the technique of image haze removal has received increasing attention in image understanding and computer vision fields during recent years.So far,the image haze removal algorithm(e.g.,algorithms for image restoration and image manipulation enhanced algorithm)based on physical model has been demonstrated to be successful.The method of single image haze removal using dark channel prior has been widely concerned(focus on the complexity and the dehazing effect).But this algorithm which estimates transmission rate unable self-adaptively with different density image,and applying classical He algorithm to image haze removal with the haze image including the sky,white cloud or the bright areas may lead to low contrast and color distortion of the image.This article mainly analyzes two shortcomings of dark channel prior algorithm in obtaining relevant parameters,and the corresponding image haze algorithm is obtained.The main works of the thesis in research are shown as following:(1)Aiming at the dark channel prior of estimate the transmission rate deficiency problems,an improved haze removal algorithm based on the mean gradient and the dark channel prior is proposed.Analyzing the dark channel priority,the transmission rate of the haze and the haze removal image is obtained.Meanwhile,the coefficient of the haze is self-adaptively adjusted by utilizing the mean gradient algorithm.Thereby,the traditional dark channel prior transmission rate is improved.The problem,which the transmission rate may not efficiently been estimated,can be solved since the coefficient of the haze is fixed.(2)An improved algorithm is proposed for image haze removal based on dark channel priority to avoid the classical He algorithm and color distortion of haze removing about sky,white cloud or bright areas.The improved algorithm is mainly reflected in the following two aspects.One is the refined atmospheric transmittance ofthe hazy image with non-sky obtained using guided filter.The other is that the atmospheric transmittance of the hazy image with the sky,white cloud or the bright areas is estimated by a variety of classified statistics for haze-free outdoor images with non-sky or the haze-free outdoor images with sky.The atmospheric transmittance of the hazy image including the sky and non-sky is obtained.(3)By the adverse foggy or hazy weather usually lose visual effect.So analyzing the feature of foggy traffic image using improved dark channel prior applying to traffic scene.And it is that the improvement of road marking features extraction,road signs detection and license plate recognition.