Efficient Hypothesis Generation through Sub-categorization for Multiple Object Detection

  • Authors:

  • Dipankar Das;Yoshinori Kobayashi;Yoshinori Kuno

  • Affiliations:

  • Graduate School of Science and Engineering, Saitama University, Saitama, Japan 338-8570;Graduate School of Science and Engineering, Saitama University, Saitama, Japan 338-8570;Graduate School of Science and Engineering, Saitama University, Saitama, Japan 338-8570

  • Venue:
  • ISVC '09 Proceedings of the 5th International Symposium on Advances in Visual Computing: Part II
  • Year:
  • 2009

Quantified Score

Hi-index 0.01

Visualization

Abstract

Hypothesis generation and verification technique has recently attracted much attention in the research on multiple object category detection and localization in images. However, the performance of this strategy greatly depends on the accuracy of generated hypotheses. This paper proposes a method of multiple category object detection adopting the hypothesis generation and verification strategy that can solve the accurate hypothesis generation problem by sub-categorization. Our generative learning algorithm automatically sub-categorizes images of each category into one or more different groups depending on the object's appearance changes. Based on these sub-categories, efficient hypotheses are generated for each object category within an image in the recognition stage. These hypotheses are then verified to determine the appropriate object categories with their locations using the discriminative classifier. We compare our approach with previous related methods on various standards and the authors' own datasets. The results show that our approach outperforms the state-of-the-art methods.