Research Article

Visible and Infrared Face Identification via Sparse Representation

Table 1

Main results. (a) Identification rates on the extended Yale B database. (b) Identification rates for each random space on the FERET database and a comparison between different normalization and fusion techniques (best score per column in bold). (c) Identification rates on the Notre-Dame database obtained by score fusion of visible and infrared modalities (best score per column in bold).
(a)

Extended Yale B database

Number of images 8 16 32
Identification rate % 88.39 95.74 98.04

(b)

FERET database
Random subspace dimension 50 100 150
Identification rate % 67.43 73.28 74.69

MM DeSc ZS MAD Tanh

PROD 68.95 77.92 77.92 77.92 78.12
SUM 78.02 73.18 78.12 78.12 78.12
MAX 77.21 67.64 0.10 0.10 75.30
MIN 68.95 73.48 74.19 74.29 74.19
78.32 64.51 78.22 78.22 78.22
76.00 13.91 74.79 74.39 74.79
78.42 73.08 78.42 78.42 78.42

(c)

Notre-Dame database
MM DeSc ZS MAD Tanh

PROD 83.12 83.12 83.12 83.12 83.12
SUM 92.32 78.52 93.17 93.18 87.41
MAX 89.86 75.40 84.99 83.99 87.41
MIN 83.12 87.41 92.82 91.99 75.47
85.86 11.19 91.80 91.15 57.77
92.83 73.50 93.02 92.11 87.41
93.47 51.71 94.06 93.88 87.99