Two-way saliency for airport detection in remote sensing images

State-of-the-art methods for airport detection in panchromatic remote sensing images utilize very limit geometrical features of airport line segments. This paper proposes a novel method based on both bottom-up and top-down saliency. Noticed that airport runways have features of vicinity and parallelity, and their lengths are among certain range, the concept of near parallelity is introduced after using a line segments detector (LSD) and a connection process. It is treated as a priori knowledge which can fully exploit geometrical relationship of airport runways to get top-down saliency. Meanwhile, an improved graph-based visual saliency (GBVS) model is used to extract bottom-up saliency. Candidate regions can be gotten by combining those two-way results. After that, scale-invariant features transform (SIFT) and support vector machine (SVM) are used to finally determine whether the regions contain airports or not. The proposed method is tested on an image dataset composed of different kinds of airports. The experimental results show that the method has advantages in terms of speed, recognition rate and false alarm rate. Also, the method is more robust to complex background.