Research On Single Fog And Haze Image Clearness Method

In recent years,most cities in China have been affected by varying degrees of haze,haze weather not only endanger human life and health,induced a variety of respiratory diseases,and seriously affect the outdoor imaging system of normal work,resulting in incalculable property damage.Image defog technology has always been a hot topic in the field of computer vision and digital image processing.It is significant to clarify the haze image.In this thesis,we have studied and improved the defog method based on model and non-model.In the non-physical model defog method,this thesis focuses on the theory of histogram equalization,the main work focused on the following two aspects:On the one hand,we analyzed the reason of grain state in the sky region when histogram equalization algorithm is applied to haze images with sky area.For the enhancement of haze image with sky area,the sky area is extracted according to the gradient and the brightness threshold,and then the independent antilog grayscale transformation is carried out,which can eliminate the grain state.On the other hand,in this thesis,an improved inverse distance interpolation algorithm is proposed to eliminate the block effect,and the smoothing control parameters are introduced to control the degree of block effect elimination.In the physical model defog method,this thesis focuses on the theory of dark channel priori,and the main work focuses on the following three aspects:First of all,Based on the idea of count sort to accelerate the acquisition of dark channel image,time complexity is reduced from O(r ~2NM)to O(rNM).Secondly,we analyzed the influence of atmospheric light intensity estimation on haze image restoration and the existing methods.The dichotomous sampling method is proposed to estimate the atmospheric light intensity.At the same time,the time complexity is reduced from O(NM)to O(log(NM)).Finally,we analyze the influence of the transmittance refinement and the defects of the guiding filter.The inherent boundary of the haze image is used to improve the normalized parameter?of guided filtering to eliminate the halo effect.In this thesis,we design and implement the image defog experiment and introduce five different objective evaluation criteria.In order to explore the inherent distinction and connection between two algorithms based on non-model and model,Experiment is carried out from two aspects:Firstly,selecting the images of different haze type in the same scene of Frida database;Secondly,using OpenGL built-in atomization model to pollute a simple scene by different intensity haze,and then carry out the clarity processing;Finally,We analyze the results of these two experiments from a subjective and objective point of view and explain the application of the scene.