Research On Color Restoration Algorithm From A Single Haze Degraded Image

Fog is a natural phenomenon due to the moisture floating in the air condenses into fine particles. Influenced by foggy weather, the images obtained by imaging device always have some characteristics low contrast, low color saturation and fuzzy scenario outline. The degradation of haze image has restricted the practical application of outdoor vision systems such as industrial measurement and monitoring, automatic navigation and unmanned reconnaissance in military aspects, which is hard to meet the requirement of the current rapid development of science and technology. So the color restoration of the fog-degraded images became one of the hotspots in the field of computer vision research. The evolution of haze removal research of a single degraded image has been slow, because of less prior information, target scene content complexity, the limitations of the degradation model itself and so on. Its research has more practical significance.The reflected light from objects reaches the imaging device after the absorption and scattering of atmospheric particles in a foggy day. The reflected light of the target scene in the process of transmission attenuates with the increase of depth of the scene. The atmospheric scattering light from non-target scene through the scattering of atmospheric particles increases with the increase of the depth, In this paper, we recover the color of a single haze image using the degradation model of fog images.Fast visibility restoration from a single color or gray level image base on median filtering is able to distinguish between fog and low saturation scenario, quickly estimate the atmospheric veil function. But the effect of dehazing in the small edge area where depth changes abruptly is incomplete. And there are excessive algorithm parameters, which is hard to get ideal results by manual Settings. To solve these problems, this paper introduced a fast restoration algorithm from a single color haze image based on image guided filter according to the atmospheric scattering model. The algorithm estimate the atmospheric veil function through two guide filtering operation to the original haze image and could have better dehazing effect at the small edge area where depth changes abruptly. It has nearly linear time complexity and less parameter. But this algorithm had poor effect in the distance. This is because there is far more noise (fog) afar to reduce the contrast, it would introduce some noise to directly mimus the edge information of the area having lower variance from the edge information of the original image. It Lead to mist removal in the distant not complete.In order to further weaken the influence of the noise in the original image, this paper improve derivation method of the atmospheric veil function on the basis of the fast algorithm using guiding filter. Firstly, the algorithm obtained the initial atmospheric veil through median filtering operation from the minimum image of each channel of the original image. Secondly, we modified the initial atmospheric veil with putting edge information obtained by guided filter as guided image. It could not only fix the initial atmospheric veil in the inaccurate edge information, but also can smooth low variance areas with noise. At the same time, we estimated air light using the theory of Dark Channel Prior to get more accurate degradated model of haze image. Finally, we recovered the color of haze image. Compared with other existing typical dehazing algorithms, this algorithm had better effect on edge information perception. It also could remove the fog in the distant scene, avoiding too much noise in the original image.