. 2021 May 15;7(5):89. doi: 10.3390/jimaging7050089

Table 2.

Characteristics of the dimensionality-reduction-based feature extraction methodologies.

Ref.	Key Features	Advantages	Disadvantages
[45]	Application of pattern map images with PCA	Fast and a high identification rate (100%)	High number of feature vectors (40 features), results depend on parameters
[46]	Application of manifold learning	Robust against pose variation, a low EER (0.80%)	Low RR (97.80%)
[47]	Combination of B2DPCA with eigenvalue normalization	Improves upon the original 2DPCA method and other methods	Low RR (97.73%)
[48]	Combination of Radon transformation and PCA	Low FAR (0.008) and FRR (0)	An in-house dataset is used instead of a benchmark one
[49]	Application of linear discriminant analysis with PCA	Very fast and retains the main feature vector	Low Accuracy (98.00%)
[50]	Application of (2D)²PCA	High RR (99.17%)	Sample increment with SMOTE
[51]	Comparison of multiple PCA algorithms	Can reach an accuracy of up to 100%	Requires a large training set
[52]	Application of KPCA	High accuracy (up to 100%)	Accuracy depends on the kernel, feature output, and training size
[53]	Combination of KMMC and 2DPCA	Improves upon the recognition time of just KMMC	Very slow recognition time
[54]	Combination of MFRAT and GridPCA	Fast and robust against vein structures, variations in illumination and rotation	Low RR (95.67%)
[55]	Application of pseudo-elliptical sampling model with PCA	Retains the spatial distribution of vein patterns, reduces redundant information and differences	High EER (1.59%) and low RR (97.61%)
[56]	Application of Discriminative Binary Codes	Fast extraction and matching with a low EER (0.0144%)	Requires the construction of a relation graph
[57]	Combination of Gabor filters and LDA	Low EER (0.12%)	Part of a multi-modal system
[58]	Application of multi-scale uniform LMP with block (2D)²PCA	Preserves local features with a high RR (99.32%)	Does not retain global features and the EER varies per dataset (high to low)