3 PFT III Datasets

Although large tests, both PFT and PFT II only used subsets of data available from the Arizona Department of Public Safety, the Los Angeles County Sheriff’s Department, and the Department of Homeland Security. For PFT III, NIST is using new subject comparisons from each of these datasets. Additionally, PFT III adds comparisons of public and sequestered data collected as part of the Intelligence Advanced Research Project Activity (IARPA)’s Nail-to-Nail (N2N) Challenge.

3.1 Arizona Department of Public Safety

The Arizona Department of Public Safety (AZDPS) dataset consists of plain and rolled impressions of all ten fingers. Figure 3.1 and Table 3.1 show the detection error tradeoff (DET) curves of all fingers not compared in other PFT tests. This data is separated by finger position in Figure 3.2 and Table 3.2 and again by impression type in Figure 3.3 and Table 3.3. Values made by combinations of fingers were generated by summing the individual similarity scores for comparisons of the individual finger and dividing by the number of values added. This technique is known as sum fusion.

Detection error tradeoff of all comparisons from all fingers in the PFT III AZDPS dataset. Numbers in gray indicate the similarity threshold.

Figure 3.1: Detection error tradeoff of all comparisons from all fingers in the PFT III AZDPS dataset. Numbers in gray indicate the similarity threshold.

Table 3.1: False non-match rate values at specific false match rates for all comparisons from all fingers in the PFT III AZDPS dataset
FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
0.0212 0.0174 0.0129
Detection error tradeoff of all comparisons from all fingers in the PFT III AZDPS dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Figure 3.2: Detection error tradeoff of all comparisons from all fingers in the PFT III AZDPS dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Table 3.2: False non-match rate values at specific false match rates for all comparisons from all fingers in the PFT III AZDPS dataset, separated by finger position. Combined finger positions were generated by sum fusion.
FRGP FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
R Thumb 0.0125 0.0108 0.0088
R Index 0.0146 0.0118 0.0083
R Middle 0.0144 0.0119 0.0092
R Ring 0.0179 0.0140 0.0101
R Little 0.0393 0.0327 0.0243
L Thumb 0.0138 0.0119 0.0097
L Index 0.0160 0.0128 0.0093
L Middle 0.0139 0.0112 0.0086
L Ring 0.0206 0.0164 0.0117
L Little 0.0488 0.0403 0.0302
R & L Thumb 0.0047 0.0040 0.0033
R & L Index 0.0038 0.0031 0.0020
R & L Middle 0.0041 0.0035 0.0026
R & L Ring 0.0053 0.0042 0.0030
R & L Little 0.0143 0.0114 0.0083
R Index & Ring 0.0053 0.0044 0.0032
L Index & Ring 0.0056 0.0048 0.0038
R Slap 0.0033 0.0028 0.0022
L Slap 0.0039 0.0033 0.0025
R Hand 0.0020 0.0017 0.0014
L Hand 0.0022 0.0019 0.0015
Detection error tradeoff of all comparisons from all fingers in the PFT III AZDPS dataset, separated by finger position and impression type. Combined finger positions were generated by sum fusion.

Figure 3.3: Detection error tradeoff of all comparisons from all fingers in the PFT III AZDPS dataset, separated by finger position and impression type. Combined finger positions were generated by sum fusion.

Table 3.3: False non-match rate values at specific false match rates for all comparisons from all fingers in the PFT III AZDPS dataset, separated by finger position and impression type. Combined finger positions were generated by sum fusion.
FRGP FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
Plain to Plain
R Thumb 0.0114 0.0096 0.0082
R Index 0.0168 0.0127 0.0095
R Middle 0.0161 0.0131 0.0097
R Ring 0.0164 0.0125 0.0095
R Little 0.0354 0.0271 0.0174
L Thumb 0.0121 0.0104 0.0087
L Index 0.0172 0.0128 0.0086
L Middle 0.0144 0.0109 0.0083
L Ring 0.0177 0.0136 0.0096
L Little 0.0505 0.0385 0.0267
R & L Thumb 0.0046 0.0040 0.0035
R & L Index 0.0034 0.0027 0.0018
R & L Middle 0.0045 0.0035 0.0027
R & L Ring 0.0044 0.0034 0.0024
R & L Little 0.0090 0.0065 0.0040
R Index & Ring 0.0057 0.0045 0.0034
L Index & Ring 0.0054 0.0047 0.0038
R Slap 0.0031 0.0028 0.0023
L Slap 0.0037 0.0034 0.0029
R Hand 0.0017 0.0014 0.0013
L Hand 0.0019 0.0017 0.0014
Plain to Rolled
R Thumb 0.0118 0.0102 0.0084
R Index 0.0138 0.0108 0.0075
R Middle 0.0138 0.0113 0.0085
R Ring 0.0176 0.0134 0.0096
R Little 0.0380 0.0303 0.0215
L Thumb 0.0130 0.0111 0.0090
L Index 0.0154 0.0120 0.0086
L Middle 0.0134 0.0105 0.0080
L Ring 0.0202 0.0157 0.0109
L Little 0.0458 0.0367 0.0260
R & L Thumb 0.0045 0.0039 0.0031
R & L Index 0.0035 0.0028 0.0017
R & L Middle 0.0036 0.0031 0.0023
R & L Ring 0.0047 0.0036 0.0025
R & L Little 0.0118 0.0092 0.0058
R Index & Ring 0.0046 0.0038 0.0029
L Index & Ring 0.0052 0.0044 0.0033
R Slap 0.0029 0.0024 0.0018
L Slap 0.0034 0.0028 0.0022
R Hand 0.0017 0.0014 0.0012
L Hand 0.0019 0.0016 0.0014
Rolled to Rolled
R Thumb 0.0149 0.0127 0.0101
R Index 0.0099 0.0079 0.0061
R Middle 0.0111 0.0095 0.0076
R Ring 0.0145 0.0123 0.0089
R Little 0.0264 0.0226 0.0179
L Thumb 0.0168 0.0146 0.0115
L Index 0.0121 0.0100 0.0080
L Middle 0.0119 0.0102 0.0082
L Ring 0.0185 0.0152 0.0119
L Little 0.0310 0.0251 0.0205
R & L Thumb 0.0047 0.0041 0.0032
R & L Index 0.0033 0.0029 0.0019
R & L Middle 0.0040 0.0036 0.0027
R & L Ring 0.0061 0.0051 0.0038
R & L Little 0.0135 0.0107 0.0089
R Index & Ring 0.0046 0.0040 0.0030
L Index & Ring 0.0050 0.0043 0.0037
R Slap 0.0032 0.0028 0.0022
L Slap 0.0037 0.0033 0.0025
R Hand 0.0026 0.0024 0.0022
L Hand 0.0027 0.0024 0.0020

3.2 Los Angeles County Sheriff’s Department

The Los Angeles County Sheriff’s Department (LASD) dataset consists of plain and rolled impressions of all ten fingers, captured with a mixture of ink and optical devices. Figure 3.4 and Table 3.4 show the DET of all fingers not compared in other PFT subsets. This data is separated by finger position in Figure 3.5 and Table 3.5 and again by impression type in Figure 3.6 and Table 3.6. Curves made by combinations of fingers were generated by sum fusion.

Detection error tradeoff of all comparisons from all fingers in the PFT III LASD dataset. Numbers in gray indicate the similarity threshold.

Figure 3.4: Detection error tradeoff of all comparisons from all fingers in the PFT III LASD dataset. Numbers in gray indicate the similarity threshold.

Table 3.4: False non-match rate values at specific false match rates for the PFT III Los Angeles County Sheriff’s Department dataset overall.
FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
0.0369 0.0301 0.0227
Detection error tradeoff of all comparisons from all fingers in the PFT III LASD dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Figure 3.5: Detection error tradeoff of all comparisons from all fingers in the PFT III LASD dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Table 3.5: False non-match rate values at specific false match rates for the PFT III LASD dataset, separated by finger position. Combined finger positions were generated by sum fusion.
FRGP FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
R Thumb 0.0152 0.0126 0.0103
R Index 0.0259 0.0199 0.0144
R Middle 0.0265 0.0222 0.0168
R Ring 0.0340 0.0276 0.0198
R Little 0.0630 0.0519 0.0398
L Thumb 0.0170 0.0141 0.0112
L Index 0.0289 0.0236 0.0173
L Middle 0.0323 0.0263 0.0203
L Ring 0.0398 0.0320 0.0237
L Little 0.0873 0.0711 0.0552
R & L Thumb 0.0036 0.0031 0.0026
R & L Index 0.0060 0.0048 0.0032
R & L Middle 0.0065 0.0052 0.0037
R & L Ring 0.0091 0.0071 0.0045
R & L Little 0.0259 0.0207 0.0147
R Index & Ring 0.0078 0.0064 0.0046
L Index & Ring 0.0106 0.0086 0.0065
R Slap 0.0041 0.0034 0.0025
L Slap 0.0062 0.0049 0.0039
R Hand 0.0020 0.0017 0.0014
L Hand 0.0022 0.0019 0.0015
Detection error tradeoff of all comparisons from all fingers in the PFT III LASD dataset, separated by finger position and impression type. Combined finger positions were generated by sum fusion.

Figure 3.6: Detection error tradeoff of all comparisons from all fingers in the PFT III LASD dataset, separated by finger position and impression type. Combined finger positions were generated by sum fusion.

Table 3.6: False non-match rate values at specific false match rates for the PFT III LASD dataset, separated by finger position and impression type. Combined finger positions were generated by sum fusion.
FRGP FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
Plain to Plain
R Thumb 0.0141 0.0115 0.0093
R Index 0.0262 0.0197 0.0138
R Middle 0.0290 0.0244 0.0183
R Ring 0.0312 0.0249 0.0187
R Little 0.0599 0.0453 0.0308
L Thumb 0.0116 0.0092 0.0065
L Index 0.0325 0.0261 0.0194
L Middle 0.0397 0.0340 0.0275
L Ring 0.0413 0.0343 0.0263
L Little 0.1076 0.0812 0.0572
R & L Thumb 0.0030 0.0026 0.0021
R & L Index 0.0046 0.0033 0.0024
R & L Middle 0.0070 0.0059 0.0043
R & L Ring 0.0063 0.0046 0.0031
R & L Little 0.0199 0.0137 0.0088
R Index & Ring 0.0069 0.0057 0.0041
L Index & Ring 0.0122 0.0104 0.0088
R Slap 0.0032 0.0028 0.0022
L Slap 0.0069 0.0064 0.0057
R Hand 0.0015 0.0014 0.0013
L Hand 0.0016 0.0013 0.0013
Plain to Rolled
R Thumb 0.0161 0.0136 0.0104
R Index 0.0321 0.0253 0.0175
R Middle 0.0357 0.0272 0.0189
R Ring 0.0444 0.0351 0.0245
R Little 0.0830 0.0688 0.0494
L Thumb 0.0192 0.0151 0.0113
L Index 0.0354 0.0270 0.0192
L Middle 0.0372 0.0296 0.0215
L Ring 0.0499 0.0385 0.0267
L Little 0.0983 0.0821 0.0587
R & L Thumb 0.0038 0.0031 0.0027
R & L Index 0.0082 0.0059 0.0039
R & L Middle 0.0076 0.0054 0.0034
R & L Ring 0.0112 0.0080 0.0046
R & L Little 0.0295 0.0229 0.0152
R Index & Ring 0.0101 0.0078 0.0053
L Index & Ring 0.0114 0.0087 0.0059
R Slap 0.0050 0.0039 0.0027
L Slap 0.0054 0.0043 0.0030
R Hand 0.0021 0.0018 0.0014
L Hand 0.0023 0.0018 0.0016
Rolled to Rolled
R Thumb 0.0149 0.0128 0.0109
R Index 0.0138 0.0098 0.0071
R Middle 0.0124 0.0106 0.0081
R Ring 0.0197 0.0153 0.0113
R Little 0.0326 0.0244 0.0176
L Thumb 0.0205 0.0179 0.0156
L Index 0.0155 0.0126 0.0090
L Middle 0.0167 0.0121 0.0092
L Ring 0.0237 0.0177 0.0129
L Little 0.0446 0.0295 0.0218
R & L Thumb 0.0042 0.0035 0.0030
R & L Index 0.0033 0.0025 0.0020
R & L Middle 0.0032 0.0025 0.0019
R & L Ring 0.0057 0.0048 0.0034
R & L Little 0.0133 0.0098 0.0070
R Index & Ring 0.0043 0.0033 0.0026
L Index & Ring 0.0048 0.0038 0.0031
R Slap 0.0021 0.0018 0.0016
L Slap 0.0026 0.0022 0.0019
R Hand 0.0015 0.0015 0.0012
L Hand 0.0020 0.0016 0.0015

3.3 Port of Entry, BioVisa Application

The Port of Entry/BioVisa Application (POE+BVA) dataset consists of plain impressions of index fingers. Figure 3.7 and Table 3.7 show the DET of all fingers not compared in other PFT subsets. This data is separated by finger position in Figure 3.8. Curves made by combinations of fingers were generated by sum fusion.

Detection error tradeoff of all comparisons from all fingers in the PFT III POE+BVA dataset Numbers in gray indicate the similarity threshold.

Figure 3.7: Detection error tradeoff of all comparisons from all fingers in the PFT III POE+BVA dataset Numbers in gray indicate the similarity threshold.

Table 3.7: False non-match rate values at specific false match rates for the PFT III POE+BVA dataset overall.
FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
0.0123 0.0093 0.0069
Detection error tradeoff of comparisons from the PFT III POE+BVA dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Figure 3.8: Detection error tradeoff of comparisons from the PFT III POE+BVA dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Table 3.8: False non-match rate values at specific false match rates for the PFT III POE+BVA dataset, separated by finger position. Combined finger positions were generated by sum fusion.
FRGP FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
R Index 0.0099 0.0078 0.0059
L Index 0.0150 0.0109 0.0079
R & L Index 0.0014 0.0012 0.0009

3.4 US VISIT #2

The US VISIT #2 (VISIT2) dataset consists of plain impressions of index fingers and are similar to POE+BVA. Figure 3.9 and Table 3.9 show the DET of all fingers not compared in other PFT subsets. This data is separated by finger position in Figure 3.10. Curves made by combinations of fingers were generated by sum fusion.

Detection error tradeoff of all comparisons from all fingers in the PFT III VISIT2 dataset. Numbers in gray indicate the similarity threshold.

Figure 3.9: Detection error tradeoff of all comparisons from all fingers in the PFT III VISIT2 dataset. Numbers in gray indicate the similarity threshold.

Table 3.9: False non-match rate values at specific false match rates for the PFT III VISIT2 dataset overall.
FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
0.0175 0.0121 0.0081
Detection error tradeoff of comparisons from the PFT III VISIT2 dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Figure 3.10: Detection error tradeoff of comparisons from the PFT III VISIT2 dataset, separated by finger position. Combined finger positions were generated by sum fusion.

Table 3.10: False non-match rate values at specific false match rates for the PFT III VISIT2 dataset, separated by finger position.
FRGP FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
R Index 0.0152 0.0111 0.0081
L Index 0.0191 0.0131 0.0081
R & L Index 0.0032 0.0025 0.0019

3.5 IARPA Nail-to-Nail

In September 2017, IARPA held a fingerprint data collection as part of the Nail to Nail Fingerprint Challenge. Participating Challengers deployed devices to capture a rolled-equivalent print. Approximately two-thirds of the ten-print data collected was released to the public as NIST Special Database 302, with the rest being sequestered at NIST for evaluations like PFT III.

Information about the Challenge can be found in NIST IR 8210. Descriptions of the devices described by the device codes in the following figures and tables can be found in NIST TN 2007.

3.5.1 By Device

The following figures and tables show the DET of all comparisons from select devices in the IARPA N2N Challenge. All probe devices imaged natively at 500 PPI, except for J, R, and U, which imaged at 1 000 PPI. The reference device, V, also imaged natively at 1 000 PPI. When these higher resolution devices are shown at 500 PPI, they were downsampled using NIST Fingerprint Image Resampler (NFIR).

Figure 3.11 and Table 3.11 show results with 500 PPI probes against 1 000 PPI references. Figure 3.12 and Table 3.12 show the same probe images against reference images downsampled to the same resolution. Figure 3.13 and Table 3.13 show native 1 000 PPI to 1 000 PPI comparisons for supported devices.

Overall detection error tradeoff of comparisons from the PFT III IARPA N2N dataset, using probe images at $500$ PPI and reference images at their native $1\,000$ PPI resolution.

Figure 3.11: Overall detection error tradeoff of comparisons from the PFT III IARPA N2N dataset, using probe images at 500 PPI and reference images at their native 1 000 PPI resolution.

Table 3.11: False non-match rate values at specific false match rates for devices deployed in the IARPA N2N Challenge at 500 PPI compared to a 1 000 PPI reference roll.
Device FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
A 0.1341 0.1088 0.0897
B 0.1593 0.1375 0.1139
C 0.0499 0.0406 0.0356
E 0.0741 0.0582 0.0482
F 0.1512 0.1231 0.0960
J 0.0497 0.0413 0.0320
K 0.0977 0.0810 0.0607
M 0.0820 0.0676 0.0510
P 0.0513 0.0391 0.0244
R 0.0284 0.0203 0.0155
S 0.0241 0.0174 0.0133
U 0.0275 0.0213 0.0175
Overall detection error tradeoff of comparisons from the PFT III IARPA N2N dataset, using probe images at $500$ PPI and reference images downsampled to $500$ PPI.

Figure 3.12: Overall detection error tradeoff of comparisons from the PFT III IARPA N2N dataset, using probe images at 500 PPI and reference images downsampled to 500 PPI.

Table 3.12: False non-match rate values at specific false match rates for devices deployed in the IARPA N2N Challenge at 500 PPI compared to a downsampled 500 PPI reference roll.
Device FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
A 0.1240 0.1031 0.0844
B 0.1560 0.1309 0.1064
C 0.0464 0.0391 0.0318
E 0.0713 0.0563 0.0448
F 0.1503 0.1167 0.0937
J 0.0457 0.0373 0.0297
K 0.0943 0.0752 0.0578
M 0.0808 0.0673 0.0513
P 0.0440 0.0220 0.0171
R 0.0257 0.0196 0.0135
S 0.0179 0.0138 0.0118
U 0.0240 0.0193 0.0143
Overall detection error tradeoff of comparisons from the PFT III IARPA N2N dataset for devices that supported native $1\,000$ PPI to $1\,000$ PPI comparisons.

Figure 3.13: Overall detection error tradeoff of comparisons from the PFT III IARPA N2N dataset for devices that supported native 1 000 PPI to 1 000 PPI comparisons.

Table 3.13: False non-match rate values at specific false match rates for 1 000 PPI devices deployed in the IARPA N2N Challenge compared to a 1 000 PPI reference roll.
Device FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
J 0.0493 0.0408 0.0328
R 0.0311 0.0257 0.0182
U 0.0289 0.0222 0.0175

3.5.2 Resample Test

PFT III supports encoding of variable resolution images. It is thought that several fingerprint feature extractors downsample imagery to a lower resolution before extracting features. To test this theory, we downsample and compare source and reference imagery both captured natively at 1 000 PPI. All downsampling was performed using NFIR.

Images were compared at all combinations of 100, 250, 300, 333, 500, 600, and 1 000 (native) PPI.

Figure 3.14 and Table 3.14 show match rates against 1 000 PPI references. Figure 3.15 and Table 3.15 show match rates against 600 PPI downsampled references. Figure 3.16 and Table 3.16 show match rates against 500 PPI downsampled references. Figure 3.17 and Table 3.17 show match rates against 333 PPI downsampled references. Figure 3.18 and Table 3.18 show match rates against 300 PPI downsampled references. Figure 3.19 and Table 3.19 show match rates against 250 PPI downsampled references. Figure 3.20 and Table 3.20 show match rates against 100 PPI downsampled references.

Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to $1\,000$ (native) images.

Figure 3.14: Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to 1 000 (native) images.

Table 3.14: False non-match rate values at specific false match rates for device U from the IARPA N2N Challenge at various resolutions compared to 1 000 (native) reference rolls from device V.
Probe Resolution (PPI) FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
100 0.8591 0.8178 0.7471
250 0.0254 0.0205 0.0167
300 0.0284 0.0240 0.0187
333 0.0278 0.0234 0.0181
500 0.0275 0.0213 0.0175
600 0.0272 0.0219 0.0170
1 000 0.0289 0.0222 0.0175
Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled $600$ PPI images.

Figure 3.15: Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled 600 PPI images.

Table 3.15: False non-match rate values at specific false match rates for device U from the IARPA N2N Challenge at various resolutions compared to downsampled 600 PPI reference rolls from device V.
Probe Resolution (PPI) FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
100 0.8556 0.8088 0.7392
250 0.0231 0.0173 0.0146
300 0.0213 0.0170 0.0140
333 0.0228 0.0187 0.0152
500 0.0246 0.0196 0.0155
600 0.0263 0.0196 0.0170
1 000 0.0251 0.0190 0.0170
Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled $500$ PPI images.

Figure 3.16: Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled 500 PPI images.

Table 3.16: False non-match rate values at specific false match rates for device U from the IARPA N2N Challenge at various resolutions compared to downsampled 500 PPI reference rolls from device V.
Probe Resolution (PPI) FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
100 0.8547 0.8038 0.7301
250 0.0237 0.0190 0.0132
300 0.0249 0.0193 0.0129
333 0.0234 0.0193 0.0143
500 0.0240 0.0193 0.0143
600 0.0246 0.0222 0.0164
1 000 0.0249 0.0196 0.0152
Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled $333$ PPI images.

Figure 3.17: Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled 333 PPI images.

Table 3.17: False non-match rate values at specific false match rates for device U from the IARPA N2N Challenge at various resolutions compared to downsampled 333 PPI reference rolls from device V.
Probe Resolution (PPI) FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
100 0.8368 0.7924 0.7137
250 0.0225 0.0181 0.0126
300 0.0216 0.0164 0.0129
333 0.0219 0.0178 0.0132
500 0.0240 0.0193 0.0152
600 0.0251 0.0193 0.0161
1 000 0.0257 0.0225 0.0170
Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled $300$ PPI images.

Figure 3.18: Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled 300 PPI images.

Table 3.18: False non-match rate values at specific false match rates for device U from the IARPA N2N Challenge at various resolutions compared to downsampled 300 PPI reference rolls from device V.
Probe Resolution (PPI) FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
100 0.8351 0.7857 0.7135
250 0.0225 0.0175 0.0143
300 0.0208 0.0167 0.0129
333 0.0219 0.0190 0.0143
500 0.0231 0.0181 0.0140
600 0.0231 0.0178 0.0143
1 000 0.0249 0.0205 0.0158
Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled $250$ PPI images.

Figure 3.19: Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled 250 PPI images.

Table 3.19: False non-match rate values at specific false match rates for device U from the IARPA N2N Challenge at various resolutions compared to downsampled 250 PPI reference rolls from device V.
Probe Resolution (PPI) FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
100 0.8281 0.7865 0.7225
250 0.0219 0.0164 0.0132
300 0.0228 0.0184 0.0123
333 0.0228 0.0178 0.0126
500 0.0231 0.0190 0.0152
600 0.0231 0.0193 0.0161
1 000 0.0251 0.0213 0.0181
Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled $100$ PPI images.

Figure 3.20: Detection error tradeoff of comparisons from the PFT III IARPA N2N dataset using downsampled probe images of various resolutions as compared to downsampled 100 PPI images.

Table 3.20: False non-match rate values at specific false match rates for device U from the IARPA N2N Challenge at various resolutions compared to downsampled 100 PPI reference rolls from device V.
Probe Resolution (PPI) FNMR @ FMR = 0.0001 FNMR @ FMR = 0.001 FNMR @ FMR = 0.01
100 0.5149 0.4503 0.3787
250 0.8401 0.7942 0.7263
300 0.8383 0.8000 0.7292
333 0.8398 0.7962 0.7211
500 0.8541 0.8132 0.7386
600 0.8541 0.8105 0.7465
1 000 0.8646 0.8216 0.7561