Study Of Haze Removal Using Single Image

With the wide use of the monitoring system, thus a new higher challenge was put forward about the availability and efficiency of the system. However, due to most of the monitoring system was installed outdoors, so the image always affected by weather. Though in cloudless day, light will also be influenced by the air. In bad weather, such as the haze weather, the situation will be more serious, in this case the image will not be clear and vivid. The reason that makes the image not clear is that the reflected light of the scene will be affected by the air more or less before it transfer to the camera, and hence the light change it intrinsic path, this situation can be more serious when the scene is far from the camera. Scattering effect of the air make it unable to use monitoring system when the weather is bad.so we need to use image processing technology to recover it so as to use monitoring system even in bad weather.Until now, there are many kinds of image recovering technology. One of them is using several images in the same scene to remove haze, this kind of method can obtain the depth information of scene easily, the process is relatively intuitive and this method is easy to realize. The more important thing is that the result is acceptable.. But the acquisition of several images in the same scene is not an easy task, so this method is not work for actual case.The other one is using single image to remove haze, this kind of methods is divided into two aspects: one is based on image enhancement and the other the one is based on atmospheric imaging model. The method based on image enhancement has developed for a long time, this kind of method has a broad application, but the haze removing result is not good enough. the method based on atmospheric imaging model can obtain good result, but it’s process is relatively complex. The above two methods has its advantages and disadvantages. In this paper, we use the two methods above to realize haze removal.The main work of this paper is below:(1) Discuss a few methods based on image enhancement, some of the classical methods above are realized with matlab, and then we give the experimental results and comparative analysis.(2) Discuss dark channel prior method haze removing method base on atmospheric imaging model, then we put forward a method based on feature recognition theory to update the calculation of atmospheric intensity. We realized this method with c++ code, and then we give the experiment results.(3) In the light of the problem of dark channel prior,we discuss two methods called automatic light region recognition and neighborhood similarity theory. We put forward a method called dark channel edge with adaptive template to improve the efficiency of the methods neighborhood similarity theory. All the above methods are realized with c++ code, and then we give the experiment results and comparative analysis.