Research And Implementation Of Image Dehazing Algorithm Under Severe Environment

Recent years, as people’s increasing demand in safety, video monitoring has been increasing widely used, so people have improved demands in the digital video monitoring system. But under severe environments, the quality of the image from video monitoring system meets great challenge. Because under severe environments, especially in the haze weather conditions, the image information often have not good visual effect, and reduced contrast and sharpness, so the application of video monitoring system is seriously affected. Therefore, it is very important to improve the degraded image quality in haze weather conditions.In this paper, we propose an improved algorithm from haze removal algorithm based on dark channel priori, and the main contributions can be summarized as follows:Firstly, when dealing with bright areas with image haze removal algorithm based on dark channel, bright areas always have poor visual effects and color distortion phenomenon. In response to this phenomenon, to improve the atmosphere light estimates value, this paper select the thickest haze as the atmosphere light estimation region, and take into account the contribution of three color channels to the luminance. The method replaces the original method, making the atmosphere light more accurate. At the same time, we add the tolerance to expand the image haze removal algorithm based on dark channel, and modify the transmission formula of the bright areas on the basis of the original mathematical model, so that we can still use this algorithm even if the haze image has large bright areas.Secondly, as the soft matting method when refining the transmission has low efficiency, guide filter is proposed to replace the original soft matting method to refine the transmission. Rough neighborhood transmission distribution image is guide filtering through guide image, that is the original hazy image, and we get the refined transmission distribution. The filtering method preserves the overall characteristics of rough neighborhood transmission distribution. At the same time, we can also obtain the variation details of the original hazy image, which solves the block effect. Guide filter uses the method of the local optimal within the window, which is different from the global optimal strategy in the soft matting. So this improved method greatly reduces the time of the original algorithm.Thirdly, noise is introduced in image haze removal process, which has a bad effect on the final results. In order to solve this problem, we use the non-local means de-noising algorithm. This method makes full use of the redundant information of the whole image to realize the weighted average de-noising, and the average of gauss neighborhood similar gray value is regarded as Pixel gray value after image de-noising. This de-noising algorithm can maintain the various details of the original features, and has good de-noising results.Finally, we propose the objective assessment methods to verify the effect of my algorithm. This paper uses a blind image quality assessment method based on property of Human Visual System. The noise distribution is weighted by the contrast sensitivity function of Human Visual System to get the objective quality assessment method, which is consistent to the subjective evaluation results. This algorithm does not require a reference image, and the computational complexity is low.The experiments show that both subjective and objective image quality assessment have verified the feasibility and effectiveness of our improved algorithm.