Skip to main content

Table 5 Average LPIPS perceptual similarity over 30000 pairs of images composed of original images and their corresponding NbNet pre-image (matching pairs) and 30000 pairs of images and pre-images generated from unrelated templates (non-matching pairs), as a function of the size of the secured templates from which the pre-images have been derived

From: Cancelable templates for secure face verification based on deep learning and random projections

Pairs

Clean

r = 64

r = 128

r = 256

Matching

0.37624

0.51812

0.50773

0.50107

Non-matching

0.52557

0.50843

0.50893

0.504596