Key Technologies Of Single Frame And Sequence Image Dehazing Based On Double Constraints

Under hazy weather conditions,the images captured by optoelectronic devices show serious image degradation problems such as contrast reduction,color distortion,blurred edges and details,which directly limit and affect the visual imaging and bring certain difficulties to the identification and judgment of targets,and even lead to its failure to work normally.Therefore,in order to reduce the influence caused by hazy weather on the visual imaging system,it is of great research significance and application value to perform fast and effective recovery processing of hazy degraded images.In this paper,the degradation mechanism of hazy image is analyzed in detail,single image dehazing,video dehazing and the objective assessment of the dehazing algorithm are thoroughly investigated,and the proposed method in the paper is proved to be effective and robust by simulation experiments.The main research contents and innovative achievements of this paper are summarized as follows:1.To address the problem of unstable consistency between the existing objective quality assessment of dehazing and subjective assessment,the differences between the traditional image quality assessment and quality assessment of the dehazing algorithm are analyzed,and the image degradation problems introduced after dehazing are summarized into three categories: poor image visibility,structural damage due to pseudo contours,and color shift.In this paper,image visibility is used to measure the change of image clarity before and after dehazing,gradient similarity and variance similarity are composed of structural similarity to measure the pseudo phenomenon,and color recovery is used to describe the color shift phenomenon.Finally,the image visibility,structural similarity and color recovery are combined to build a new image dehazing algorithm quality assessment system.The experimental results and data analysis show that compared with the typical image quality assessment methods,the proposed dehazing algorithm quality assessment system performs better in terms of rank correlation,linear correlation and prediction accuracy,and is more consistent with the subjective assessment.2.For the limitations and defects of the dark channel prior in single image dehazing,a double constrained dehazing algorithm based on local continuity and global boundary is proposed.The algorithm firstly combines the scene radiation constraint and piecewise smoothness constraint to establish the energy function of atmospheric transmission from a new perspective,and adopts the ?-expansion technique for energy minimization.The optimized transmission can avoid the block effect and prevent the pixel value overflow in the dehazed image.Secondly,an adaptive edge-aware weight is constructed based on the guided filtering with exponential function and edge information to suppress the halo effect.Finally,to avoid the interference brought by bright objects in the scene to the atmospheric light estimation,the bright pixels in the image are properly corrupted by an improved non-overlapping dark channel and a quadratic tree hierarchical search is used to improve the accuracy of atmospheric light estimation values.Experimental results and data analysis show that the proposed algorithm can recover haze-free images with more realistic colors and higher scene clarity,and better suppression of visual artifacts compared with typical dehazing algorithms.3.To address the lack of spatio-temporal coherence problem commonly existing in current video sequence dehazing methods,this paper proposes a video sequence dehazing method based on double constraints of temporal and spatial information.The method maintains the temporal coherence between the adjacent frames by the temporal correlation term,and constructs a spatial correlation term to suppress the cumulative error of long video sequences and enhance the spatial consistency of the dehazed video.Based on the pixel differences of adjacent frames,a weighting function is constructed using a probabilistic model,and the weighting function is used to correct the temporal correlation term and spatial correlation term to solve the situation of large differences in the scene depth information of adjacent frames.Atmospheric light is corrected by updating the formula to further ensure the color consistency between adjacent frames.In addition,the integral map algorithm and the downsampling technique are used to accelerate the algorithm to ensure the real-time requirements of the algorithm in engineering applications.The experimental results and data analysis show that the proposed method can better suppress the flicker effect and visual artifacts compared with the typical video dehazing methods,and can ensure the real-time performance of the video algorithm,which provides fundamental technical support for engineering applications.