Abstract: While the dark channel prior algorithm works well for single image haze removal, it has to take long time to process and require expensive computational overhead and huge memory resource. In this paper, a fast daze removal algorithm is proposed based on dark channel prior knowledge. In the proposed approach, the structure of dark channel filter is designed to estimate refined transmission rate instead of the original soft-matting operator, which will substantially reduce the computational complexity. To deal with the invalid case of the proposed method, the histogram specification is used based on original image to enhance brightness and optimize gray distribution of haze-free image, which improves the visual effect of defogging images. Compared with He’s algorithm, the proposed approach is equaled to the original algorithm in daze removal effect, and has about five times shorter processing time.