Abstract Due to the serious threats to road safety caused by pavement distress, it has drawn much attention to seek efficient road crack detection methods for automatic pavement condition evaluation. In recent detectors, structure information of cracks has not been utilized effectively, which leads to the high sensitivity to crack-like noises. In this paper, we propose a novel crack detection method to recognize cracks out of noises. We verified our method in a real pavement crack dataset. Extensive experiments demonstrate the state-of-the-art results of the proposed method.