In this paper, we present a new tone mapping technique for high dynamic range images based on the retinex theory. Our algorithm consists of two steps, global adaptation and local adaptation of the human visual system. In the local adaptation process, the Gaussian filter of the retinex algorithms is substituted with a guided filter to reduce halo artifacts. To guarantee good rendition and dynamic range compression, we propose a contrast enhancement factor based on the luminance values of the scene. In addition, an adaptive nonlinearity offset is introduced to deal with the strength of the logarithm function's nonlinearity. Experiments show that our algorithm provides satisfactory results while preserving details and reducing halo artifacts.