Linfeng Xu, Liaoyuan Zeng, Zhengning Wang. Saliency-based superpixels. Signal, Image and Video Processing, 2014.

Xu et al. propose a saliency-based method for generating oversegmentations based on superpixels. However, I want to note that the used notion of oversegmentation (or superpixels) is slightly different in that the authors explicitly want individual objects to be covered by few superpixels. Therefore, the proposed method can also be seen as general segmentation algorithm. Without going into detail, their approach merges superpixels based on saliency. Specifically, the algorithm has similarities to [1] in that edges between pixels (and, thus, between superpixels) are assigned saliency values. Superpixels are then merged only if the created segment increases the overall saliency. Qualitative results are shown in Figure 1.

Figure 1: Qualtiative results in comparison to selected superpixel algorithms and graph-based segmentation [1].

  • [1] P. Felzenswalb and D. Huttenlocher. Efficient graph-based image segmentation. IJCV, 2004.
What is your opinion on this article? Let me know your thoughts on Twitter @davidstutz92 or LinkedIn in/davidstutz92.