Research On Image Haze Removal

With the development of science and technology become more and more advanced. Multimedia image processing technology has been widely used in various fields. But the camera influenced by various factors, such as fog, rain and smoke which cause the sharpness of the image is reduced, serious can cause the image color information offset or distortion. In order to make monitoring system be able to use in various weather and can normal work. Therefore, the system must include some mechanism to remove the influence of the poor weather which has very important practical significance. The research in this issue focused on haze removal and did following researches:This paper researches the reasons of fog and the atmospheric scattering model. And studies the advantage and disadvantage of haze removal based on Retinex theory.Based on the center surround Retinex an improved algorithm is proposed. Because of the fog reduce the contrast of the image, so before the Retinex algorithm using an improved histogram equalization to adjust the brightness of global image, which in order to increase the contrast.Using tanh instead of log,because tanh has better dynamic range compress and brightness ability.As a result of the traditional MSR is just the sum of each subband, then the high frequency subband together many times that will cause noise pollution. Therefore, we use each subband phase subtraction to avoid this problem.Using an adaptive space varying gain function to instead of constant gain. The result of different scales in each channel has different gain which will lead to the detail of the image is more prominent and will effectively curb the phenomenon of the halo.The above method applied to the video image and presents two improvements. On the one hand because of the fog mainly affect the low frequency part of image, so we just only to handle the low part of the image. On the other hand Gaussian filter takes up most of the time, therefore the recursive Gaussian filter is used.Combined with the experimental results and theoretical analysis, this algorithm has achieved a better visual effect.