Abstract: In order to solve the problems of the dark channel prior algorithm, such as color oversaturation, overall luminance darkening and halo in the sky, an image dehazing algorithm fusing the luminance model and gradient domain filter is proposed. Firstly, the average value of the first 0.1% pixels with the largest luminance is selected as the atmospheric light value. Secondly, the improved dark channel model and luminance model of adaptive minimum filter are used to solve the transmission in the foreground region and the sky region respectively, and then the rough transmission  is obtained by weighted fusion. On the basis of that, the gradient domain guided filter is used to refine the transmission. Finally, the haze-free image is restored by atmospheric scattering model and gamma correction. The experimental results show that this algorithm can quickly and effectively solve the halo effect and image distortion of the sky region on the haze image containing the sky region, and the restored image is clear and natural, with more detailed information retained, which is better than other comparison algorithms in subjective and objective evaluation. 

Key words: image dehazing, dark channel prior, luminance model, gradient domain guided filter, atmospheric scattering model