Object detection by admissible region search

  • Authors:

  • Xiaoming Chen;Senjian An;Wanquan Liu;Wanqing Li

  • Affiliations:

  • Department of Computing, Curtin University, Perth, Australia;Department of Computing, Curtin University, Perth, Australia;Department of Computing, Curtin University, Perth, Australia;Information and Communication Technology (ICT) Research Institute, University of Wollongong, NSW, Australia

  • Venue:
  • AI'11 Proceedings of the 24th international conference on Advances in Artificial Intelligence
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Efficient Subwindow Search(ESS) is an effective method for object detection and localization, which adopts a scheme of branch-and-bound to find the global optimum of a quality function from all the possible subimages. Since the number of possible subimage is $\emph{O}(\emph{n}^{4})$ for an images with $\emph{n}\times\emph{n}$ resolution, the time complexity of ESS ranges from $\emph{O}(\emph{n}^{2})$ to $\emph{O}(\emph{n}^{4})$ . In other words, ESS is equivalent to the exhaustive search in the worst case. In this paper, we propose a new method named Adimissible Region Search(ARS) for detecting and localizing the object with arbitrary shape in an image. Compared with the sliding window methods using ESS, ARS has two advantages: firstly, the time complexity is quadratic and stable so that it is more suitable to process large resolution images; secondly, the admissible region is adaptable to match the real shape of the target object and thus more suitable to represent the object. The experimental results on PASCAL VOC 2006 demonstrate that the proposed method is much faster than the ESS method on average.