Figure 1.
Each examiner was assigned a distinct, randomized sequence of latent-exemplar image pairs. For each pair, the latent was presented first for a value decision. If the latent was determined to be no value, the test proceeded directly to the latent from the next image pair; otherwise, an exemplar was presented for comparison and evaluation.
Figure 2.
Example annotation of a mated image pair.
Corresponding features are indicated here in red, unassociated features in yellow, and debatable correspondences with question marks. This examiner marked 8 corresponding minutiae, 2 debatable correspondences, individualized, and assessed it as very difficult. Determinations by the 9 examiners assigned this image pair: 5 NV, 4 VID (1 changed to NV during Comparison); 1 inconclusive, 2 individualization.
Table 1.
Design dimensions used for data selection.
Figure 3.
Associations of (A) minutia count and value determinations from analysis of the latent (n = 3730); (B) corresponding minutia count and determinations from comparison of latent and exemplar prints on mated data (n = 2796).
In (B), 1.6% of determinations with 12 or more corresponding minutiae marked were not individualized. A few responses in (B) indicate NV with corresponding minutiae due to examiners changing their value determinations during Comparison.
Table 2.
Misclassification rates for models describing associations between annotations and individualization determinations by the same examiner.
Figure 4.
Example of a mated image pair (A), showing variations in annotation among five examiners (B–F).
Corresponding points are shown here in red, unassociated in blue; minutiae as circles, deltas as triangles, other points as rhombuses; noncorresponding points as red Xs. Examiners B–E individualized; F excluded. Determinations by the 11 examiners assigned this image pair: 2 NV, 3 VEO (2 of which were changed to VID during Comparison), 6 VID; 1 inconclusive, 1 exclusion, 7 individualization.
Figure 5.
Corresponding minutia count (y-axis) and determination (color) by image pair (x-axis).
Each column of points contains the set of all responses for a given image pair. Some points are superimposed, indicated through color blending. X-axis is sorted by median, then by mean corresponding minutia count. Latents that were determined NV and not compared are shown as having zero corresponding minutiae. NV responses with one or more corresponding minutiae are due to examiners changing their value determinations during Comparison. (n = 2796 responses by 165 examiners to 231 mated image pairs.)
Figure 6.
Detail of Figure 5 for the 39 image pairs that had median corresponding minutia counts between 6 and 9.5, with the addition of box plots showing interquartile range, minima, and maxima.
(n = 452 responses; 6 to 16 responses per image pair.)
Table 3.
Image pair and examiner effects on corresponding minutia counts, showing restricted maximum likelihood estimates.
Table 4.
Misclassification rates for models using one examiner's annotations and determinations to predict a second examiner's individualization determinations.
Figure 7.
Logistic models estimating the probability of individualization based on corresponding minutia counts, on mated image pairs: (green) probability that an examiner would individualize based on the same examiner's corresponding minutia counts (6.0% misclassification, see Table 2); (red) probability that another examiner would individualize based on this examiner's minutia counts (20.4% misclassification, Table 4); (blue) probability that an examiner would individualize based on the median of all examiners' corresponding minutia counts (13.6% misclassification, Table 5).
Table 5.
Misclassification rates for models describing individualization as a dependent response to (A) image pairs and examiners and (B) attributes of the image pairs as estimated by median statistics (derived from all examiner responses).
Figure 8.
Distribution of corresponding minutia counts by (A) the majority of participants (n = 2062 comparisons of mated pairs by 155 examiners) and (B) those participants following a 12-point standard (n = 135 comparisons of mated pairs by 10 examiners).
Colored by determination: inconclusive (black), individualization (gray); NV not included.