Graphics Reference
In-Depth Information
To verify the effectiveness of the proposed features extraction method for infrared
face recognition, the four existing features extraction algorithms are used for
comparisons that include PCA+LDA [13], DCT+LDA [7], DCT+ FS+LDA [8]. We
have used our own implementations of all of these algorithms on our infrared face
database. The best performances of different algorithms are shown in Table 1.
Table 1. Best recognition rates of different algorithms
Methods
Results
DCT+PLS (Proposed)
95.8%
PCA+PLS
92.4%
DCT+FS +LDA[8]
93.6%
DCT+LDA[7]
91.2%
PCA+LDA[13]
89.2%
It is revealed from tabe1 that the recognition performance of the algorithm based
on DCT and PLS is very high and outperforms that of the methods based on DCT and
LDA. This is because the PLS is a powerful feature extraction technique for
discriminative information in DCT domain. We observed that the PLS regression
performed better than LDA because PLS basis projects the feature vectors into a
latent space in which feature vectors corresponding to the same subject are closer than
the feature vectors corresponding to different subjects.
5
Conclusions
In this research paper we presented a DCT based feature extraction method for the
representation of infrared face images. To perform face recognition, the proposed
features were classified using the PLS regression. The experiments were performed
on our infrared face datasets and the results of the proposed algorithm were compared
with other state-of-the art infrared face recognition algorithms based on DCT and
PCA. The experimental results proved that the proposed algorithm consistently
outperforms the existing methods.
Acknowledgements. While working on this research paper, we were supported by the
National Nature Science Foundation of China (No. 61201456), the Natural Science
Foundation of Jiangxi Province of China (No. 20132BAB201052), the Science &
Technology Project of Education Bureau of Jiangxi Province (No.GJJ14581) and the
Nature Science Project of Jiangxi Science and Technology University
(2013QNBJRC005, 2013ZDPYJD04); we would like to show our highest respect here
to all those who provided help.
 
Search WWH ::




Custom Search