In this paper we present multiple methods to augment a graph-based foreground detection scheme which uses the smallest nonzero eigenvector to compute the saliency scores in the image. First, we present an augmented background prior to improve the foreground segmentation results. Furthermore, we present and demonstrate three complementary methods, which allow for detection of the foregrounds containing multiple subjects. The first method performs an iterative segmentation of the image to "pull out" the various salient objects in the image. In the second method, we used a higher dimensional embedding of the image graph to estimate the saliency score and extract multiple salient objects. The last method, using a proposed heuristic based on eigenvalue difference, constructs a saliency map of an image using a predetermined number of smallest eigenvectors. Experimental results show that the proposed methods do succeed in extracting multiple foreground subject more successfully as compared to the original method.
Michal Kucer, Nathan D. Cahill, Alexander C. Loui, David W. Messinger, "Augmenting Salient Foreground Detection using Fiedler Vector for Multi-Object Segmentation" in Proc. IS&T Int’l. Symp. on Electronic Imaging: Computational Imaging XV, 2017, pp 116 - 121, https://doi.org/10.2352/ISSN.2470-1173.2017.17.COIMG-433