Fast Video Defogging Algorithm Based On Subsampling

Atmospheric visibility reduces in bad weather due to fog present in the atmosphere, Images taken under such bad weather conditions suffer from degradation and severe contrast loss. With the degradation of the environment and increased frequency of foggy weather, the defog algorithms have important implications for the outdoor activities risk reduction and enhancement of vision on foggy days. In this thesis a fast defog algorithm is proposed for video to expand the application of image defog algorithms.After the research of relevant algorithms, we made a comparison between defog algorithms based on image enhancement and defog algorithms based on physical model of atmospheric scattering, and demonstrated the superiority of the latter. Next we analyzed the image degradation caused by fog with the help of atmospheric scattering model, and further verified the feasibility of atmospheric scattering model’s application in defog algorithm.Among those defog algorithms based on atmospheric scattering model, multi-image dehazing algorithms can get parameters in advance only with the application to more than one photographs or the help of other auxiliary tools,failing to satisfy real-time defogging, while their application is limited due to sensitivity to environmental changes; single-image dehazing algorithms can achieve parameters only through complicated procedures, slowing down the speed of fog removal. We compared the two kinds of algorithms, finding that single-image defogging algorithms are more suitable for video defog with optimized procedures and parameter estimation.A fast single-image defogging algorithm was proposed in this thesis, to achieve simplified algorithm procedure, we figure out the second variant in the atmospheric scattering model (atmospheric environment light) instead of the whole transmission function; we proceed two simple filtering instead of detailed transmission image; we also reduced the amount of calculation and speed up the procedure with the help of subsample algorithm. We did some experiments on the algorithm of this thesis and other powerful algorithms, it turns out that this thesis’s algorithm have competitive performance in variants estimation and video defog, and the algorithm has better performance in algorithm complexity and computational efficiency.