Background priors based saliency object detection

Zexia Liu, Guanghua Gu, Chunxia Chen, Dong Cui, Chunyu Lin

Abstract: Saliency object detection is the key process of identifying the location of the object. It has been widely used in numerous applications, including object recognition, image segmentation, video summarization and so on. In this paper, we proposed a saliency object detection approach based on the background priors. First, we obtain a border set by collecting the image border superpixels, in addition remove the superpixels with strong image edges out of the border set to reduce the foreground noises and obtain the true background superpixels seeds. Then, the initial saliency map can be made by computing a background saliency map based on the background seeds and fusing a centered anisotropic Gaussian distribution. Finally, we refine the initial saliency map via the smoothness constraint which encourages neighbor pixels in the image to have the same label. Experimental results on two large benchmark datasets demonstrate that the proposed algorithm performs favorably against other six state-of-art methods in terms of precision, recall and F-Measure. Our method is demonstrated to be more effective in highlighting the salient objects and reducing the background noise.