Abstract: Based on the assumption of spatially smooth illuminant,traditional center surround-based Retinex enhancement methods suffer the existence of halo artifacts.In this article,we propose a fast Retinex image enhancement method that models global and local adaptation of the human visual system.A global dynamic range adjustment is performed first.Then mean shift filtering on the image is carried out in the illuminant estimation step,and the spatial-varying illuminant is eliminated base on the center surround hypothesis.Outliers at both extremities of the histogram are removed in the post-processing step.Experimental results demonstrate that our method efficiently renders high dynamic range images and reduces halo artifacts.An accelerated implementation of mean shift is employed,thus our method is much faster than previous methods.