Research On Image Defogging Algorithms Based On Atmospheric Scattering Model

In a real natural environment,the visual effect of foggy images captured outdoors is often severely degraded.This phenomenon is caused by tiny particles in the air such as water vapor,smoke particles,etc.which interfere with the propagation of light.At the same time,the reflected light from the target scene is attenuated before it enters the camera lens.The light ingested by the camera is a mixture of scattered ambient light and attenuated reflected light from the scene,which results in images obtained by the image acquisition device showing an overall reduction in visual impacts,reduced contrast,and the presence of large areas of greyish white.The existence of fog also limits the in-depth research of many computer vision applications such as target detection,image classification,face recognition,etc.Therefore,it is of great importance to research image-dehazing algorithms.Based on the principle of the atmospheric scattering model,we further analyze and study the causes of foggy images,and finally propose two single image dehazing algorithms to achieve the clarity of foggy images by using some image processing methods and optimization algorithms in machine learning such as Gamma Correction,Fibonacci algorithm,Tone Mapping,etc.The main innovative works are as follows.1.Based on the analysis of existing atmospheric scattering models,a real-time image dehazing algorithm based on gamma correction is proposed.We found that there is a linear mapping relationship between hazy and haze-free images,and this linear mapping relationship is only determined by the transmittance of the scene according to the experiment.In this thesis,we take advantage of the fact that gamma correction can effectively fit multiple linear mapping functions and combine some inherent properties of natural images to constrain and adjust the shape of the gamma correction function.Firstly,the global atmospheric light A is estimated using a modified quadtree segmentation algorithm and a normalization of the fog map is done by using the estimated atmospheric light,then the minimum channel of the pseudo-dehazed map and the corresponding pseudo-transmittance map is estimated using the specially designed gamma correction function in this thesis.Afterward,the Fibonacci method with image downsampling is used to determine the optimal transmittance map.Finally,a clear fog-free image is obtained according to the recovery formula of the atmospheric scattering model.Extensive experiments have demonstrated the method's effectiveness and low complexity.2.To address the limitations of the dark channel prior(DCP)theory,a fast dark channel dehazing algorithm based on multiscale weighted guided image filtering is proposed in this thesis.Firstly,an improved quadtree algorithm is used to determine the atmospheric light value and normalize the image;then,the minimum channel of the fog map is morphologically blurred to obtain a preliminary estimate of the atmospheric light layer(Veil),while the atmospheric light layer is refined using our proposed adaptive atmospheric light layer compensation strategy,then an initial transmittance map is obtained;secondly,the multi-scale weighted guided image filter combined with a resizing algorithm based on image pyramid is used to refine the transmittance map.Finally,a clear image is recovered based on the atmospheric scattering model,and the brightness of the output is selectively fine-tuned using a tone mapping algorithm.Through extensive experiments and comparisons with other dehazing algorithms in terms of subjective visual effects and objective metrics,the algorithm is shown to have excellent performance and timeliness.