Guided image filter is one of the most commonly used ways to refine transmission maps. However, since this filter transfers the structures of the guidance image to the filtering output, when the guidance image is the input image itself, even small textures in the input image will cause the change of transmission, which is obviously contrary to the principle that transmission changes only when scene depth changes. In this paper, saliency detection, which simulates the way human eyes work, is introduced into haze removal to tackle the above issue. We first use saliency detection to capture the depth change regions, and then the saliency value is used as an adjustable factor to compute proper guidance images, in which most texture details are blurred but the depth change regions are remained clearly visible. Experimental results show that our method has great superiority in detail recovery compared with other state-of-art methods.