Instance-Level landmark labeling via multi-layer superpixels

  • Authors:

  • Yanyun Qu;Jiangjun Yang;Han Liu;Yi Xie;Cuihua Li

  • Affiliations:

  • Computer Science Department, Xiamen University, P.R.China;Computer Science Department, Xiamen University, P.R.China;Computer Science Department, Xiamen University, P.R.China;Computer Science Department, Xiamen University, P.R.China;Computer Science Department, Xiamen University, P.R.China

  • Venue:
  • PCM'12 Proceedings of the 13th Pacific-Rim conference on Advances in Multimedia Information Processing
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Millions of place-specified photos are uploaded on the Internet. Landmark labeling is very important for place-specified image understanding, landmark retrieval and auto-annotation. In this paper, we aim at extracting and labeling a Landmark in an image. The novelty of our method is that we use multi-layer superpixels to effectively extract a Landmark. The multi-layer superpixels can be used to capture the context of scale space and the spatial coherency of neighboring superpixels. And the context constraints are enforced by Conditional Random Field. In our method, we firstly learn a SVM classifier which operates on the superpixels of the training data. Then we construct a 3D adjacent graph which links the superpixels not only in the same layer but also in the successive layers. Finally, we use Conditional Random Field to combine the supervision information with the context cues in order to label landmarks. We compare our method with the state-of-the-art methods on the landmark images which are collected from Flickr, and the experimental results show that our method has achieved the best detection precision and the best pixel-based precision-recall.