Research Of Image Enhancement Algorithms In Extreme Weather Conditions

Images of inclement weather, such as haze, rain, snow and dust often have severe degradation. The image contrast is reduced,details are blurred and so on. These degraded images directly limit the application of traffic surveillance,target recognition and navigation which need clear input images. Therefore, it is very important to make the degraded image of inclement weather clear. This paper studies the reason of degradation, proposes the corresponding image enhancement algorithms of different weather combined with the physical model of degraded images.In the aspect of haze image enhancement processing, this paper studies the improved dark channel prior algorithm which is based on the dark channel theory. According to the haze image model,the relevant parameters can be solved and then the real scene of the image can be acquired. This algorithm can eliminate the influence of the haze on the image,but when there are large bright areas in the image, the dark channel theory fails.Images enhanced by this method would appear color distortion. This paper puts forward the corresponding improvent aiming at disadvantages of the dark channel prior: 1.The dark channel prior theory is a statistical rule, it fails in images which contain a large scale of bright areas. In this paper, the proximity of the input image and the airlight value is used as the threshold to judge the bright area. The transmission of the bright region is corrected by the fast estimation algorithm proposed in this paper, while the other regions are solved by the original algorithm. 2. The estimation of the airlight by dark channel prior can not exclude the interference of large white part of the near scene. The value of the airlight affects the acquisition of the transmisssion directly. In order to get more accurate transmission, this paper uses the local entropy method to select the airlight value in the most hazy area, which avoides the interference of white objects effectively. After the improvement, the contrast and the clarity of the restored image are improved, the color is natural and the color distorrion is revised.In the aspect of rain and snow image enhancement processing,first,an adaptive bilateral filter is applied to the input image to maintain the edge information and filter out the influence of the noise. Second,after studying the rain and snow image model, according to the similarity of this model and the haze image model, the improved haze image enhancement algorithm based on the dark channel prior is applied to the rain and snow image enhancement. The rain and snow are removed and the clarity of the image is strengthened.In the aspect of dust image enhancement processing, this paper compares the RGB histograms of haze and dust images.The histograms of haze images are consistent while the histograms of dust images are not, because most of the blue light is absorbed by the atmospheric particles in dust images. This paper analyzes the cause of color cast in dust image and rectifies it to make the image consistent with the true color of the scene. Finally, the improved dark channel prior method is applied to enhance the corrected image. The influence of the dust is removed and the clarity of the image is improved.