Weak metric learning for feature fusion towards perception-inspired object recognition

  • Authors:

  • Xiong Li;Xu Zhao;Yun Fu;Yuncai Liu

  • Affiliations:

  • Institute of Image Processing & Pattern Recognition, Shanghai Jiao Tong University, Shanghai, China;Institute of Image Processing & Pattern Recognition, Shanghai Jiao Tong University, Shanghai, China;Department of CSE, University at Buffalo (SUNY), NY;Institute of Image Processing & Pattern Recognition, Shanghai Jiao Tong University, Shanghai, China

  • Venue:
  • MMM'10 Proceedings of the 16th international conference on Advances in Multimedia Modeling
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

With extracted local features of a given image, computing its global feature under perceptual framework has shown promising performance in object recognition. However, under some tough applications with large intra-class variance, using only one kind of local feature is inadequate to build a robust classification system. To integrate the discriminability of complementary local features, in this paper, we extend the efficacy of perceptual framework to adapt to heterogeneous features. Given multiple raw global features, we propose a fusion strategy through metric learning, which is called weak metric learning in this work, for fusing high dimensional features. The fusion model is solved with the maximal kernel canonical correlation formulation with the multiple global features as outputs. Experimental results show that our method achieves significant improvements about 5% to 11% than the benchmark perceptual framework system, HMAX, on several difficult categories of object recognition with much less training samples and feature elements.