In this paper, we propose an improved method for single image haze removal based on the dark channel prior. The dark channel prior is a kind of statistics of outdoor haze-free images. It estimates the transmission map and gives out a superior result. While refining the transmission map with soft matting is computation consuming and this prior is invalid for near white scenes. We simplify the refining step by locally processing and the region to be refined is located by dealing with the coarse dark channel. By computing the difference of intensities between the highest channel and the lowest, we reestimate the transmission map with a method derived from underwater dehazing. Results show that our method has a good ability of haze removal with lower computation and the output of near white scenes is more realistic.