A Study On Image Dehazing Algorithm Based On Dark-light Channel Prior And Self-adaptive Parameter Optimization

The image information collected by outdoor imaging equipment has been missed due to the frequent occurrence of haze in recent years.This causes great inconvenience to the subsequent computer vision tasks,which seriously affects the normal operation of security monitoring systems,satellite imaging systems,traffic monitoring systems and many other systems,and brings huge security risks to people's lives.Therefore,the study on image dehazing has very important practical significance.At present,the mainstream method of dehazing based on dark channel prior mainly has the following problems: it is not applicable to large-area white regions,the estimated values of atmospheric light are too high,dehazing parameters are relatively single and post-dehazing darker images.This paper has effectively improved these issues.In the process of improvement,the main innovation points of this paper are as follows:1.Against the problem that dark channel prior is not applicable to large-area white regions,this paper proposes a new dehazing algorithm based on dark-light channel prior.According to the feature that the pixel values in the white region are generally higher,the theory of bight channel prior is proposed,which combines the theory of dark channel prior,thus the problem of dehazing for the white regions in foggy images could be effectively solved.2.To solve the problem that existing atmospheric light values tend to be too large,this paper proposes a self-adaptive weighted method for obtaining atmospheric light values.This method can make the obtained atmospheric light values more robust by adaptively weighting the pixel values and the maximum dark channel values in sky-like region.3.Against the problem that the ? weight that can decide dehazing degree is relatively single when transmittance is estimated by the existing dehazing algorithm,this paper proposes an image dehazing algorithm that can optimize the adaptive weights.Through a large number of experiments that were made using this algorithm,it was found that there was a certain relationship between the ? weight and the atmospheric light value at the time of dehazing,and then the optimized processing for the post-dehazing images was achieved perfectly based on this relationship.4.Against the problem that the filtering window scale of the existing guided filter algorithm is relatively single,this paper proposes a guided filtering algorithm with selfadaptive window scale.According to the size of the original foggy image,the algorithm can self-adaptively adjust the filtering scale so that the filtered effect is better.In this paper,after the original algorithm was improved,a combination of subjective and objective evaluations were used,a variety of dehazing algorithms were selected,in order to compare and evaluate the algorithm of this paper.The results show that the proposed algorithm not only effectively solves the shortcomings such as the original algorithm is not suitable for large-area white regions and post-dehazing darker images,but also makes the post-dehazing visual effects more realistic and natural.