Saliency-based localising active contour for automatic natural object segmentation

In this study, a novel method named saliency-seeded active contour is presented for automatic natural object extraction. Since approximately the location of the desired object can easily be obtained by saliency regions or pixels in the map, we propose the maximum saliency density method to detect salient object pixels in spite of the cluttered background at first. Then, the salient object pixels are employed as the seeds of convex hull to generate the initial contour for our automatic object segmentation system. It is most important that the method proposed by the authors does not require considerable user interaction in contrast with localising region-based active contours (LRACs), that is, the segmentation task is fulfiled in a fully automatic manner. Extensive experiments results on a large variety of natural images confirm that the framework can reliably and automatically extract the object from the complex background.