Single-image haze removal using the mean vector L2-norm of RGB image sample window

Poor visibility in bad weather is a major problem for many applications of computer vision. This paper presents an efficient single RGB image haze removal algorithm to improve image visibility. The proposed algorithm has three parts. Firstly, we estimate the initial transmission map, which is key to single-image haze removal. Different form dark channel-based method, the proposed method estimates this map by computing the mean vector L2-norm of the sample window. Then, the algorithm refines the estimated transmission map using a guidance filter. Secondly, we estimate the atmospheric light using the RGB value of the pixel corresponding to the maximum value of the L2-norm of the sample windows. Thirdly, the proposed algorithm improves the recovery approach for clear image restoration. This means that it can directly recover sufficiently bright images. Experimental results show that four parameters containing the radius of the sample window for the transmission map estimate, the guidance filter radius for refining the transmission, and the threshold and coefficient used to restore the clear image have important effects on the defogging results. Compared with the dark channel-based method, the proposed method results in brighter and more natural defogged images.