Figure 1.
Flowchart of the proposed algorithm.
Figure 2.
Feature matching before cropping the palm region near the wrist.
Figure 3.
(a) Original images; (b) Cropping the palm region near the wrist; (c) DoG filtering; (d) Histogram equalization.
Figure 4.
Feature extraction before preprocessing (Left) and after preprocessing (Right).
Figure 5.
SIFT ((a), (b), (c)) and RootSIFT ((d), (e), (f)) are compared for their robustness against rotation and scale transformations.
Figure 6.
Robustness of SIFT and RootSIFT against rotation and scale transformations.
(a) Robustness of SIFT and RootSIFT against rotation transformation; (b) Robustness of SIFT and RootSIFT against scale transformation.
Table 1.
Robustness of SIFT and RootSIFT against rotation and scale transformations.
Figure 7.
Matching results of the RootSIFT algorithm.
(a) Intra-group matching; (b) Inter-group matching.
Figure 8.
Flow chart of LBP-based mismatching removal.
Figure 9.
(a) Intra-group matching; (b) Inter-group matching.
Figure 10.
(a) Far from the camera lens; (b) Close to the camera lens; (c) Tilts to the left; (d) Tilts to the right; (e) Tilts forward; (f) Tilts backward.
Figure 11.
EER curves for different approaches.
Table 2.
Comparative experiments for two different databases.
Table 3.
The time consumption for different methods (s).
Table 4.
Verification results from the CASIA multi-spectral palmprint database.
Figure 12.
Examples of five preprocessing methods.
(a) Weber illumination normalization; (b) Retinex based illumination normalization; (c) Background estimation; (d) Our method on ROI; (e) Our method.
Figure 13.
Comparative experiment using five different preprocessing methods.
(a) EERs of the five preprocessing approaches for different distRatio values; (b) EER curves for the five preprocessing approaches for an optimal distRatio.
Figure 14.
Results from different methods for mismatching removal.