Research On Fog Removal Methods Of Video Surveillance Images

Current video and image systems are typically of limited use in the area of video surveillance, face recognition and intelligent vehicle surveillance for their poor visibility and contrast caused by the presence of fog, haze and mist. In order to provide clear images for image processing applications, different dehazing methods have been proposed in this paper according to the characteristics of video surveillance. This paper gives a deep research in dehazing from two aspects based on the analysis of image degradation, including image restoration and image enhancement.First of all, fog image degradation model is studied in this paper, and close analysis of image processing is given. Model verification is implemented by eliminating natural illumination, verifying transmittance and computing model error. In view of some factors that affect the model can be neglected under specified weather conditions, two methods of model simplification are presented in this paper.Secondly, two novel fast haze removal algorithms based on depth prior is proposed:fog image restoration from multiple images of the same depth and different depths in uniform bad weather conditions which base on the atmospheric scattering model. And the comparison with typical algorithms are showed. Experimental results demonstrate that the proposed algorithms remove haze effectively and achieve accurate restoration in video surveillance images.Then an improved haze removal algorithm is proposed based upon polarization principle. The work principle of polarizer is studied and the polarization difference equation is noticed that has a similar form with the dichromatic atmospheric scattering model. So a novel algorithm that polarized images combined with dichromatic atmospheric scattering model is proposed. On the other hand, a new method is proposed by integrating two fundamental methods grounded on the existing work that has been done. Then a necessary comparison and analysis between the proposed method and other methods are presented. Experimental results verified the effectiveness of this approach.Finally, an image enhancement algorithm based on range image segmentation is proposed. In order to verify the objects in one segmented image are in the same depth, the algorithm begins with the invariance of iso-depth neighborhoods to weather conditions, and obtains the range image by using the dark channel prior. Then the k-means algorithm is adopted to segment and gray scale transformation is applied to the final segmentation results respectively.The algorithms proposed in this paper have different application conditions, and both the theoretical analysis and experimental results demonstrate that all of them have satisfactory effects. Meanwhile, the depth information is acquired without using any expensive distance measuring equipments.