Person-Specific SIFT Features for Face Recognition

Jun Luo, Yong Ma, Erina Takikawa, Shihong Lao, Masato Kawade, Bao-Liang Lu

Abstract

ScaleInvariant Feature Transform (SIFT) proposed by Lowehasbeenwidely andsuccessfully applied toobject detection andrecognition. However, therepresentation ability ofSIFTfeatures infacerecognition hasrarely been investigated systematically. Inthis paper, weproposed to usetheperson-specific SIFTfeatures andasimple nonstatistical matching strategy combined withlocal andglobal similarity onkey-points clusters tosolve facerecognition problems. Largescale experiments onFERETandCASPEALfacedatabases using onlyonetraining sample per person havebeencarried outtocompare itwithother non person-specific features suchasGaborwavelet feature and LocalBinary Pattern feature. Theexperimental results demonstrate therobustness ofSIFTfeatures toexpression, accessory andposevariations. IndexTerms-SIFT, person-specific, face recognition
BibTeX
@inproceedings{luo2007person,
  author = {Jun Luo and Yong Ma and Erina Takikawa and Shihong Lao and Masato Kawade and Bao-Liang Lu},
  title = {{Person-Specific SIFT Features for Face Recognition}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2007},
  doi = {10.1109/icassp.2007.366305},
}

Loading related content...

Figure