In this paper, we propose a dehazing algorithm based on the dark-channel image centroid offset. The algorithm clusters the dark channels of hazy images to divide these images into scenes. Further, it analyzes and calculates the centroid offset of the dark-channel image of each scene to correct the transmission rate of the scene. Combined with the quadtree search algorithm, an atmospheric light estimation method based on the depth of field step image is proposed, which enables the estimated position of atmospheric light to fall in a region with a large depth of field without being affected by white or flat objects. The experimental results reveal that the proposed algorithm can effectively restore the original hue of bright regions as well as the detail information. Moreover, the restored images have appropriate brightness and natural color. Subjectively, the restored images have relatively good visual effects. Objectively, the evaluation indexes of the restored images by the proposed algorithm are overall better than those by the dark-channel-prior algorithm.