Night Image Haze Removal Based On Joint Optimization Of Multi-feature And Edge-Detail Preserving

Night images captured in the weather of fog and haze will suffer from some problems such as non-uniform illumination,color distortion and low contrast,which seriously affect the performance of video surveillance and target recognition in computer vision system.Based on the characteristics of hazy images at night,the methods of estimating ambient light and transmission is proposed,and the problems of color distortion,detail loss and noise amplification is also solved after haze removal.The main work is as follows:Due to the influence of artificial light source,the ambient light distribution in night images is non-uniform,the existing methods can easily misestimate ambient light as point light source,which is not suitable for night images.In this paper,the most relevant low frequency components of haze concentration are obtained by extracting incident light from night scene.The method of local ambient light estimation based on low-pass filtering is proposed,which is more suitable to night scenes with artificial light sources.Then,an effective transmission estimation method is presented based on the joint optimization of multi-feature which combined contrast,saturation and information entropy in order to address the inapplicability of daytime hazy image priors.Finally,as for the color deviation,the non-overlapping blocking is applied to local Shade of Gray to correct the color of the resulted image.Experimental results demonstrate that the proposed method can significantly remove haze,improve the contrast and recover more natural color.The traditional haze removal method based on atmospheric curtain has poor edge,which cause the image edge details lost.In addition,most haze removal method do not consider noise.In order to remove haze and preserve edge details of the image as much as possible,and also remove noise,the image is decomposed based on the total variation model,this method can remove the haze from the structure layer to ensure that the texture information is not destroyed.In the process of haze removal,a multi-scale fusion method is proposed to estimate ambient light,which can effectively remove halo in restored images.In order to restore more edge details,a multi-directional weighted total variation regularization method for atmospheric curtain estimation is proposed.Finally,the noise of texture layer is removed adaptively.Experimental results show that the proposed method can significantly remove haze,recover more edge details and reduce noise.