A man-made object area extraction method based on visual saliency detection and graph-cut segmentation for high resolution remote sensing imagery

Object detection and extraction are very important research topic in remote sensing processing and analysis. An object-oriented based accurate object extraction method was proposed by combining saliency detection and image segmentation. Firstly, a new saliency detection method which is adequate for high resolution remote sensing image analysis is presented by fusing graph-based visual saliency detection and line density visual saliency detection. By introducing line density, the proposed method can detect building regions under very complex background remote sensing images in an unsupervised manner. Then, graph-cut based segmentation is used to obtain image regions. Pixels in each region have similar saliency scores and features. Accurate boundaries of objects can be extracted by analyzing saliency of these regions. Compared with pixel based salient objects detection methods, our method has high true detection rate as well as low false detection rate by using object-oriented idea. Experimental results also demonstrate that our method can detect human buildings accurate target boundary.