Nighttime image defogging is critical for various applications, such as nighttime video surveillance, and target identification. The existing defogging algorithms for nighttime image have many shortcomings, such as high image distortion, fuzzy detail and poor robustness. In order to solve these problems, a new nighttime image defogging algorithm based on the theory of Retinex and dark channel prior is proposed combining with the characteristics of atmospheric scattering model and nighttime image of fog. Firstly, the incidence image affected by haze and the reflection image affected by haze are obtained based on the theory of Retinex. Then, the clear reflection image is obtained by the dark channel prior. After that, the position of the source and the depth of field are obtained according to the nighttime image of fog and the reflection image affected by haze. The sum of the distances between the spot and the source is obtained by using the camera imaging mechanism, and then the fog-free incident light image is obtained. Finally, the theory of Retinex is used to restore the nighttime clear image. Experimental results show that the proposed algorithm not only defogs completely and enhances the image contrast, but also reduces the color distortion greatly.