A Gabor-Block-Based Kernel Discriminative Common Vector Approach Using Cosine Kernels for Human Face Recognition

Table 2

Specificity and sensitivity measure of the FRAV2D dataset.

Total no. of classes = 100, total no. of images = 1800
		Individual belonging to a particular class
		Using first 3 images of an individual as training images
		Positive	Negative
FRAV2D test	Positive	𝑇𝑃 =759	𝐹𝑃=7
	Negative	𝐹𝑁=41	𝑇𝑁=693
		Sensitivity=𝑇𝑃/(𝑇𝑃+𝐹𝑁)≈94.875%	Specificity=𝑇𝑁/(𝐹𝑃+𝑇𝑁)=99.0%
		Using first 4 images of an individual as training images
		Positive	Negative
FRAV2D test	Positive	𝑇𝑃=678	𝐹𝑃=2
	Negative	𝐹𝑁=22	𝑇𝑁=698
		Sensitivity=𝑇𝑃/(𝑇𝑃+𝐹𝑁)≈96.85%	Specificity=𝑇𝑁/(𝐹𝑃+𝑇𝑁)≈99.7%

So considering the first 4 images in Figures 2(a)–2(d) of a particular individual for training the achieved rates are as follows.
False positive rate = 𝐹_𝑃/(𝐹_𝑃 + 𝑇_𝑁) = 1 − Specificity = .3%.
False negative rate = 𝐹_𝑁/(𝑇_𝑃 + 𝐹_𝑁) = 1 – Sensitivity = 3.15%.
Accuracy = (𝑇_𝑃 + 𝑇_𝑁)/(𝑇_𝑃 + 𝑇_𝑁 + 𝐹_𝑃 + 𝐹_𝑁) ≈ 98.3%.
So considering the first 3 images in Figures 2(a)–2(c) of a particular individual for training the achieved rates are as follows.
False positive rate = 𝐹_𝑃/(𝐹_𝑃 + 𝑇_𝑁) = 1 − Specificity = 1%.
False negative rate = 𝐹_𝑁/(𝑇_𝑃 + 𝐹_𝑁) = 1 – Sensitivity = 5.125%.
Accuracy = (𝑇_𝑃 + 𝑇_𝑁)/(𝑇_𝑃 + 𝑇_𝑁 + 𝐹_𝑃 + 𝐹_𝑁) ≈ 96.9%.