Research On Image Dehazing Algorithm Based On Atmospheric Scattering Model

In severe weather such as haze,a large amount of particulate matter is suspended in the atmosphere.Due to the scattering and absorption of these suspended particles,the captured images have problems such as quality degradation,blurred details,and low recognition.It brings great difficulties to image processing,computer vision,artificial intelligence and other related fields,and seriously affects the application and development of these fields.Therefore,research on image dehazing technology has important significance and value,and it is urgent to propose an effective image dehazing algorithm to restore the clarity and visibility of hazy images.Many existing dehazing methods have phenomena such as color distortion,blocking effect,and halo.By studying the main reasons for image quality degradation in hazy weather,this paper proposes an image dehazing algorithm based on dark channel prior to use nearinfrared(NIR)information to identify sky regions,and another image dehazing algorithm that combines residual blocks and dense blocks of encoder-decoder network to efficiently achieve image dehazing.And through quantitative evaluation and qualitative evaluation,compare and analyze the dehazing effect of the algorithm proposed in this paper and other dehazing algorithms.The main work and conclusions of the paper are as follows:(1)The dark channel prior dehazing algorithm is the most typical dehazing algorithm,but it is not suitable for processing sky areas.The wavelength of near-infrared light is longer than that of visible light,making it more penetrating,less affected by the scattering of suspended particles in the air,and carrying more detailed information.This paper first uses the near-infrared information recognition algorithm to distinguish the sky area and the non-sky area in the image;then,based on the dark channel prior dehazing algorithm,the improved atmospheric light estimation and the corrected sky area transmittance estimation are used to increase the accuracy of the atmospheric light value and transmittance;finally,the image dehazing is achieved through the image restoration model.Experimental results confirm that the algorithm proposed in this paper has obtained satisfactory dehazing results in solving problems such as sky areas and darker scenes.(2)The emergence and use of deep learning algorithms have greatly promoted the development of image processing.Each pixel value in the depth map reflects the true distance of the sensor from the object.Therefore,there is a more important relationship between the haze and the depth map.Based on the existing algorithms,this paper first uses deep super-resolution algorithm to process the training set NYU2 Depth Database to improve the resolution of the depth map to achieve the optimization of the training set;then combines with the Convolutional Neural Network(CNN)technologies such as residual network and dense connection network,design and build encoder-decoder network;finally,uses the optimized training set to train the encoder-decoder network to improve the learning ability of the network model and avoid losing important image feature information.The experimental results confirm that the dehazing results obtained by using the dehazing algorithm proposed in this article are more accurate in haze removal and more natural in terms of vision.