Saliency Detection via Graph-based Manifold Ranking

Chuan Yang¹, Lihe Zhang¹ , Huchuan Lu¹, Xiang Ruan², and Ming-Hsuan Yang³

¹ Dalian University of Technology, ²OMRON Corporation , ³ University of California at Merced

Figure 1.  From top to bottom: original image; saliency map and ground truth

Abstract

Most existing bottom-up methods measure the foreground saliency of a pixel or region based on its contrast within a local context or the entire image, whereas a few methods focus on segmenting out background regions and thereby salient objects. Instead of considering the contrast between the salient objects and their surrounding regions, we consider both foreground and background cues in a different way.  We rank the similarity of the image elements (pixels or regions) with foreground cues or background cues via graph-based manifold ranking. The saliency of the image elements is defined based on their relevances to the given seeds or queries. We represent the image as a close-loop graph with superpixels as nodes. These nodes are ranked based on the similarity to background and foreground queries, based on affinity matrices. Saliency detection is carried out in a two-stage scheme to extract background regions and foreground salient objects efficiently. Experimental results on two large benchmark databases demonstrate the proposed method performs well when against the state-of-the-art methods in terms of accuracy and speed. We also create a more difficult benchmark database containing 5,172 images to test the proposed saliency model and make this database publicly available with this paper for further studies in the saliency field.

Paper

Saliency Detection via Graph-based Manifold Ranking

Chuan Yang, Lihe Zhang, Huchuan Lu, Xiang Ruan, Ming-Hsuan Yang

Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2013), Portland, June, 2013.[PDF][CODE]

Comparison with other methods

Figure 2.  Left, middle: precision-recall curves of different methods. Right: precision, recall and F-measure using an adaptive threshold. All results are computed on the MSRA-1000 dataset. The proposed method performs well in all these metrics.

Figure 3.  Left: precision-recall curves of different methods. Right: precision, recall and F-measure for adaptive threshold. All results are computed on the MSRA dataset.

Figure 4.  Left: precision-recall curves of different methods. Right: precision, recall and F-measure for adaptive threshold. All results are computed on the DUT-OMRON dataset.

Downloads

DUT-OMRON: [Link]

Matlab code: The implementation of our method. [Code] (Tested on Windows with MATLAB R2010b)

Saliency results: The saliency maps of our method on the MSRA-1000, MSRA and DUT-OMRON databases. [MSRA-1000Map]  [MSRAMap] [DUT-OMRONMap]

If you have any questions, please contact:

Lihe Zhang, zhanglihe at mail dot dlut dot edu dot cn

Or

Huchuan Lu, lhchuan at mail dot dlut dot edu dot cn

Or

Minghsuan Yang, mhyang at ucmerced dot edu
　

-----------------------------------------------------------------

Chuan Yang Last Updated: 2013-03-28 . . .