Object retrieval using configurations of salient regions

  • Authors:

  • Weijun Wang;Yupin Luo;Guangrong Tang

  • Affiliations:

  • Tsinghua University, Beijing, China;Tsinghua University, Beijing, China;Tsinghua University, Beijing, China

  • Venue:
  • CIVR '08 Proceedings of the 2008 international conference on Content-based image and video retrieval
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, we describe an approach for retrieving a ranked list of images containing the same object as the query. Images in the corpus are primitively represented by a set of salient region descriptors so that the retrieval approach can works despite of changes in scale, illumination, and partial occlusion. Based on the sparse frequency vector representation method, we have improved the bag-of-features method to integrate the spatial co-occurrence information into the image representation. Instead of using the spatial configuration as a further verification stage following a prior filter stage, our method integrates the spatial information into the image representation and merges two stages in order to speed up the retrieval performance. To efficiently use the spatial configurations of salient regions, we propose a practical method to explore the content of image and flexibly cluster the salient regions into groups of neighbours. And the combination of information from the local co-occurrence of salient regions with the sparse frequency vector representation method is a major contribution of our work. Experiment results on a ground-truth dataset and complexity comparison are provided to demonstrate the advantage of our way of using spatial information for object retrieval work.