Multiscale Feature Fusion for Face Identification

Xin Wei, Hui Wang, Huan Wan, Bryan Scotney

Research output: Contribution to conferencePaperpeer-review

2 Citations (Scopus)


Over the past 20 years, information fusion has seen rapid development and has been used in many applications, including pattern recognition and computer vision. Multi-modal biometrics is one form of information fusion where biometric information from different modes of input (e.g., finger print, iris image, face image) is fused in order to achieve superior recognition performance. In this paper we propose a novel approach to multiscale information fusion for face identification where facial information from a single mode of input (i.e. a single image) at different spatial scales of focus is fused. We consider Local Binary Pattern (LBP) image descriptor, and carefully choose the LBP parameters to decide the spatial scale. Thus different values of the LBP parameters lead to LBP feature vectors at different spatial scales. These different LBP feature vectors are `fused' through an image distance function, which is then used for face image classification. We call this fusion method Multiscale Feature Fusion (MFF). Extensive experimentation has been conducted on the ORL, AR and FERET face image databases. Results show MFF does indeed produce superior classification performance when compared with the best results reported in the literature.
Original languageEnglish
Publication statusPublished (in print/issue) - 20 Jul 2017


  • Face
  • Feature Extraction
  • Iris Recognition
  • Fuses
  • Image Representation
  • Histograms


Dive into the research topics of 'Multiscale Feature Fusion for Face Identification'. Together they form a unique fingerprint.

Cite this