Research On Single Image Dehazing Algorithm Based On Non-local And Dark Channel Prior

In inclement weather such as haze and fog,large particles floating in the atmosphere will disturb the spread of light,which will cause the optical imaging system to be disturbed during the imaging process,make the image appear low contrast and insufficient color saturation,which will cause vast problems to the subsequent processing research of computer vision such as target tracking,motion recognition,and pedestrian detection.Moreover,the haze has great impact on a series of outdoor optical imaging devices such as automatic driving of cars,security monitoring,and smoke detection based on computer vision,even severe interferes the normal work of these systems,which brings huge hidden danger and harm to people.Therefore,it is of great research significance and practical value to estimate the original ideal image from the images disturbed by haze,improve the imaging quality of outdoor imaging systems in haze weather,and reduce the hidden dangers caused by haze to people's travel.In order to effectively improve the image quality affected by haze,this paper starts from the characteristics of haze and the derivation of atmospheric scattering model,then studies in detail the influence of haze weather on the optical imaging system and the cause of the decline of image quality due to by haze.Based on the analysis and summary of the popular image dehazing algorithms.Then the problems in the dark channel prior and non-local prior image dehazing algorithms are discussed and summarized in detail,and a lot of effective solutions are proposed for these problems in this paper.In brief,the contributions and work of this paper focus on the following aspects:1.The non-local prior image dehazing algorithm has improved the estimation error due to the limitations of the k-means clustering algorithm in atmospheric light estimation.After a detailed description of Neutrosophy and its application in image processing and cluster analysis,the fuzzy c-means clustering algorithm based on Neutrosophy is used instead of the k-means clustering algorithm to fit haze lines in RGB space and estimate the value of atmospheric light.In order to solve the problems of halo effect and color shift caused by the non-local prior image dehazing algorithm due to the absence of the relationship between image pixels,this paper proposes a transmittance optimization model in combination with the initial transmittance obtained from the dark channel prior.In this model,not only the non-local prior's transmittance and the transmittance of the dark channel are fused,but also the image denoising item and the transmittance denoising item of the restored image are included to suppress the problem of noise amplification.Experiments show that compared with the traditional image dehazing algorithm,the non-local prior image dehazing algorithm based on neutroscopy has more accurate atmospheric light estimation ability,image restoration ability and better visual effects.2.In order to solve the problem of wrong estimation of atmospheric light in the sky,the lamp and other extremely bright regions,a method of estimation of atmospheric light base on Yin-Yang-Sine and Cosine algorithm is used in this paper.The principle and steps of Yin-Yang pair algorithm and Sine and Cosine algorithm are described in detail,then a Yin-Yang-Sine and Cosine algorithm is proposed.Moreover,the OTSU segmentation algorithm is optimized by using this algorithm.The dark channel of the haze image is divided into seven parts.The part with the second brightness is selected to calculate its gray histogram.In order to reduce the halo phenomenon,this paper has carefully studied the characteristics of the halo in the transmittance,and based on these characteristics,a new transmittance optimization model,which combine with the general generalized total variational algorithm,is proposed.This model can obtain satisfactory restored images based on problems such as suppressing halo,removing noise,and protecting textures.Experiments show that the dark channel prior image dedehazing algorithm based on Yin-Yang-Sine and Cosine algorithm proposed in this paper can obtain high-quality restored images.3.According to the traditional dark channel prior,color attenuation prior and non-local prior,and the non-local prior based on neutroscopy and the dark channel prior image dehazing algorithm based on the yin-yang-sine cosine algorithm,this paper develop an image defogging and evaluation system base on MATLAB's Graph User Interface tool.The system includes several modules such as file operation,image defogging,manual adjustment and image quality evaluation.In particular,the application of manual adjustment module and restoration image quality evaluation module makes the system not only helpful in the study of image defogging,but also quite universal.