Image Dehazing Algorithm Based On Prior Knowledge

Under the hazy weather condition,fog particles in the air have a negative impact on propagation of light rays,resulting in photo shoots with low visibility,low contrast,which brings great inconvenience,like image segmentation,target tracking,behavior detection,to the follow-up computer vision tasks,directly affects the proper functioning of the existing outdoor imaging system(such as security and protection monitoring system)and brings huge hidden troubles to people's production and life.Therefore,it is of great practical significance to study how to improve the quality of fog-degraded image restoration results and reduce the adverse effects of haze on the existing outdoor imaging system.Based on the characteristics of hazy weather,this paper analyzes the influence of haze particles on the image process and the degradation process of fog-degraded image.Through in-depth analysis and research of the existing image de-fogging algorithm,it is found that shortcomings and improvements exist.Then from the atmospheric optical scattering model and through theoretical derivation,this paper proves the existing problems and puts forward the improvement method and at the same time improves and perfects the existing de-fogging algorithm in obtaining atmospheric light.The main works and innovation of this paper include the following aspects:1.Aiming at the problem of color distortion under dark channel prior de-fogging algorithm,a new improved method is presented to calculate the transmittance of all channels respectively.This paper firstly analyzed the influence of incident light frequency on the transmittance of all color channels on the basis of Lambert-Beer's law and elicited the ratio among the transmittance of all channels;carried out a down sampling pretreatment on images to acquire refined transmittance and then improved the efficiency of the algorithm through restoring the original size;finally obtained the transmittance of all color channel through proportional relation and employed restored images of transmittance on each channel respectively.The experimental results show that compared results of the improved image de-fogging algorithm with the existing algorithm,the image color calculated by the improved algorithm is more natural.The improved algorithm eliminates the drawback of high color saturation brought by the existing algorithm,and shorten the running time of four to nine times.2.The existing de-fogging algorithm has a problem of rough estimation of atmospheric light,which leads to a large error in the estimation result and causes color distortion in terms of restored images.To solve the problem,this paper puts forward a method of atmospheric optical estimation based on cluster statistics.First of all,select some of the potential atmospheric optical source in the original image and set a clustering on the source through the clustering algorithm to obtain several alternative clusters of atmospheric optical source;then carry on the statistics on points in each cluster and calculate the atmospheric light with the cluster that have most points;finally take brightness mean vector of the alternative atmospheric optical points in each cluster as the estimation value of the atmospheric light and the geometric center of each point on the image as the position of the atmospheric light.The experimental results show that the atmospheric brightness vector and the position of the light source obtained through cluster statistics are more accurate,which makes the image restoration results look more natural in the subjective vision and greatly enhance the objective quality evaluation index of all kinds of images.3.The existing method carries out clustering with a fixed number of alternative atmospheric optical points and estimates the atmospheric light with the cluster statistics that have most alternative points.Due to the small number of alternative point samples,the estimated atmospheric light has a large error in the statistical sense.To solve this problem,this paper uses threshold segmentation method to select alternative atmospheric optical points so as to improve the quantity of alternative atmospheric optical point samples.At the same time,ant colony algorithm is used to cluster the cluster of atmospheric light to improve the accuracy of the estimation results.Also,to improve the efficiency of the algorithm,this paper firstly carries on the preliminary clustering of alternative atmospheric optical points with K-means algorithm,and then uses ant colony algorithm to improve the clustering results.The experimental results show that the atmospheric light obtained by this method is more natural,and can further improve the image quality evaluation index.4.The existing de-fogging algorithm usually assumes that the atmospheric light of the whole image is constant and uniform.On the contrary,in many realistic situations,the atmospheric light usually changes with the depth of the scene.To solve this problem,this paper presents an algorithm of atmospheric light estimation based on a Gaussian distribution.In this algorithm,the alternative points are selected by threshold segmentation to increase the number of initial samples;meanwhile,the clustering algorithm is used to merge the cluster of source points obtained by the original algorithm to improve the number of sample points;use proportional threshold to filter out the unreasonable clusters;at the same time regard each cluster as a separate source to calculate its effect on the surrounding pixels separately;the effect is modeled by two-dimensional Gaussian distribution function;replace global atmospheric optical restoration image with atmospheric optical map.The experimental results show that the results of the restoration of the atmospheric optical image by Gauss distribution are more natural in subjective vision than the results obtained by the original algorithm,and are improved on the objective image quality evaluation index.