Novel Approaches to Improve Iris Recognition System Performance Based on Local Quality Evaluation and Feature Fusion

Table 1

Comparison of different methods.

Methods	Iris database	Feature extraction methods	Evaluation methods	Conclusions
Du et al. [3]	CASIA-V1	1-D Log Gabor	Accuracy rate	Inner rings of iris have more distinguishable and individually unique signal, and a partial iris image can be used for human identification
Hollingsworth et al. [5]	ICE (NIR)	1-D Log Gabor,; 2D Gabor	Hamming distance	Difference in consistency of iris code bits based on the size of filter used. The middle bands may be slightly better than either the inner or the outer bands
Pereira and Veiga [6]	CASIA-V1	Genetic algorithm	Hamming distance	The points that were selected non-uniformly over the iris region are more reliable than those selected uniformly
Bolle et al. [7]	—	—	—	Explained that not all bits are equally to flip but some bits are fragile bits
Yuan and Shi [11]	—	Quadrature Log-Gabor	Hamming distance	Inner has abundance of minute texture, middle has bigger block of texture, and outer is the flattest of the iris texture
Broussard et al. [1]	UBIRIS	Directional energy filter	Accuracy rate	About 40 percent of the iris produced the highest accuracy; the left and right iris produce higher identification accuracy than the top and bottom iris
Poursaberi and Araabi [13]	CASIA-V1	Wavelet Daubechies2	Hamming and harmonic mean distance	Observed that relying on a smaller but more reliable part of the iris can improve the overall performance
Proposed	CASIA-V1,; CASIA-V3 Interval,; MMU-V1,; JLUBRIRIS-V1	2D Gabor,; GLCM	SVM,; KNN,	Different iris regions on the recognition accuracy will produce different effects; several tracks close to pupil contain more feature information; track0 has much less identification accuracy than the middle tracks do; partial iris can not completely replace the entire iris