Research On Image Defogging Algorithm Under Complex Lighting

With the development of technology,computer vision system has been greatly developed,playing an important role in people's daily life,such as vehicle navigation,drone shooting,video surveillance,intelligent transportation.Most of the information come from images and videos.A majority of visual devices could capture clear and natural images without external interference.However,under the conditions of fog and haze,the reflected light will be significantly attenuated when propagating due to the scattering and absorption of atmospheric media.This leads to varying degree of degraded image received by the outdoor image acquisition device,which greatly affects and limits the normal performance of the visual system.Therefore,how to obtain high-definition images becomes more important.The research of image processing algorithms under foggy complex lighting conditions has important practical application value and theoretical research significance.Based on the principles and technologies of classic image defogging algorithms,and taking the atmospheric scattering model as the physical model,an improved image defogging algorithm under complex lighting was proposed.The main research contents are as follows:(1)In view of the problems of serious color deviation and color distortion after defogging in bright image areas in dark channel priori theory,an improved dark channel prior image dehazing algorithm is proposed.To begin with,a more accurate global atmospheric light intensity value is obtained by using the three-channel bright area segmentation method.What is more,the transmissivity image is obtained through boundary constrain conditions,and the image is smoothed using Gaussian homomorphic filtering.Finally,the transmissivity image and original image are integrated through color compensation principle and wavelet transform.Then the fused image and the defogged image are fused and compensated for multiple times to obtain the best color image,which improves the problem of color distortion and halo effect after defogging.(2)The improved multi-scale convolutional neural network is proposed against the problems of poor defogging effect ascribed to the inaccurate estimation of convolutional neural network transmissivity.First of all,the image in foggy day is decomposed into high and low frequency sub-image through bilateral filter,and convolution kernels of different sizes are used to extract the image characteristics in foggy days of high and low frequency.Furthermore,these foggy features are fused,and the maximum pooling method is combined to make the high-frequency information in the estimated transmittance more retained.The nonlinear mapping relations between the fused characteristics and transmissivity are learned through the fully connected layer.The stimulation experiment result shows that compared with other algorithms,the improved algorithm is superior in terms of feeling the various image restoration quality evaluation indicators from the perspective of subjective vision.