Research On Image Dehazing Method Based On Dark Channel Prior

At present,computer vision acquisition systems are increasingly widely used in intelligent transportation,aviation reconnaissance,electric power inspection,terrain survey,and agricultural production.However,in the harsh environment,such as fog and haze,there are a large number of particles suspended in the air,Which causes scattering during the propagation of the reflected light from the scene.This leads to the image quality obtained by the acquisition device to be degraded to varying degrees,which seriously affects the accuracy of subsequent image recognition and tracking.Therefore,how to restore the clear original appearance scene by defogging and improve the image quality is a research hotspot.This paper studies the method of defogging based on the dark channel prior.Aiming at the problems of the dark channel prior algorithm after defogging processing in the sky and other areas with sudden depth of field changes,such as color shift,artifacts,low contrast,and insufficient defogging.This paper proposes a defogging algorithm based on atmospheric scattering model estimation.Firstly,according to the logarithmic transformed and adaptive parameters,the dark channel of the fog-free image is obtained,and the atmospheric scattering model is deformed to obtain the initial transmittance,and estimated the initial transmission map by atmospheric scattering model.The difference between pixel brightness and saturation compensates the defogging strength of the sky and other depth areas.The optimized transmittance is obtained after L1 regularization.Secondly,according to the relationship between the gray-scale pixels of the haze image,the three pixel local areas can be clustered,and then best atmospheric value light for each local area is estimated by quadtree algorithm.Finally,the optimized transmittance and local atmospheric light value are used to obtain a defogging image through the atmospheric scattering model.The simulation results show that the proposed algorithm can better improve the color distortion and artifacts at the edge of the image depth of field,so that the restored image has rich details and moderate brightness.Compared with other 5 dehazing algorithms,It is averaged increased by about 7.76% on objective parameters such as peak signal-to-noise ratio,average gradient,structural similarity,and information entropy,the overall visual effect of the dehazing image is good.For the information of the dark channel prior algorithm relies on,there are certain constraints.Moreover,there are many model parameters,and the structure and procedures is complicated and cumbersome,which greatly limits the scope of scene adaptation.Aiming at the problems of the dark channel prior algorithm after defogging processing,such as color shift,loss of depth and edge details.According to the idea of extracting feature pixels of the dark channel,this paper proposes a dark channel convolutional network image defogging algorithm,combining the dark channel feature layer with Maxout activation function enhancement and multi-scale parallel feature layer to extract more haze image features to ensure the integrity of image information.And then,using residual network with hybrid dilated convolution and deep feature fusion method to improve the accuracy and richness of the detailed information between the feature layers of the haze image.The high-dimensional information is mapped from the feature space back to the image RGB space by the nonlinear regression reconstruction layer.In the end,this paper uses the target loss function to adjust the network parameters,so that the error between the predicted image and the actual clear image is continuously reduced,and the paper realize the image defogging.The simulation results show that the improved algorithm has moderate color fidelity after defogging,rich texture details,and little influence from noise.Compared with other 5 dehazing algorithms,It is averaged increased by about 9.14% on objective parameters such as peak signal-to-noise ratio,average gradient,structural similarity,and information entropy,the overall visual perception of the restored image is clear and natural.According to actual engineering application requirements,this paper designs defogging system based on the improved two different defogging algorithms,and a visual example test is carried out in a real haze environment.The results show that the improved defogging algorithm in the paper can improve the image quality and lay a good foundation for subsequent image processing.