Abstract:

Salient object detection is an important tool for highlighting the foreground region accurately, however most traditional algorithms do not work well when dealing with complex background images. We thus propose a novel salient object detection method based on foreground enhancing and background suppressing. For acquiring multiple superpixel regions, we segment the source image by the simple linear iterative clustering (SLIC), and the salient regions and the seed background of the image can be obtained via region contrast and boundary information, based on which the regioncontrastbased and backgroundbased salient images are achieved. Then we distinguish salient regions from nonsalient regions in the two salient images using the seam carving and graphbased image segmentation, which is helpful for obtaining the foreground enhancing and background suppressing template. Furthermore, we integrate the salient images and template as the final salient image. Experimental results which are applied to a public benchmark dataset (MSRA1000) show that the proposed algorithm has better precision and recall ratio than the seven classical algorithms.

Key words: salient object detection, super-pixel, foreground enhancing, background suppressing