Abstract: In this psper a novel nighttime image dehazing model based on the full-scale Retinex algorithm is proposed. We firstly estimate the global transmissivity of the image based on the dark prior theory. Then the estimated transmissivity values are mapped to the full-scale surround functions to get the illumination component. Finally, Retinex principle is used to dehaze. This method can be used to deal with various nighttime under foggy and inhomogeneous illumination conditions effectively, and to get full-scale transmissivity of the image.