Salient region detection in high resolution remote sensing images

Abstract

In this paper, we address the problem of automatic pre-segmentation for object detection and recognition in remote sensing image processing. It plays an important role in reducing computational burden and increasing efficiency for further image processing and analysis. A visual-attention based saliency computation approach is introduced to select the perceptually salient and highly informative regions that represent the main contents of the high resolution remote sensing images. In our method, two bottom-up visual saliency computation methods, edge-based and Graph-based visual saliency (GBVS), are adopted to exploit different kind of features, and the two saliency maps are fused using a 2D Gaussian shaped function for the purpose of improving salient region detection performance. The experimental results demonstrate that our proposed method performs well in ground-truth evaluation and outperforms on the salient target area segmentation task, thus could be introduced for preprocessing of targets object detection and recognition.