Fig 1.
An example of realistic fingerprint casework taken from the National Institutes of Standards and Technology Special Database 27.
This example demonstrates the noisy and partial nature of a latent print (left) compared to the matched inked print (right).
Fig 2.
Close up of an inked print from the dataset used by Busey et al.
[35]. Shown are the locations of the core and delta used in level one analysis and examples of the two fundamental minutiae—ridge endings and bifurcations—that are used in level-two analysis.
Fig 3.
A prototypical scanpath of a novice (a) and an expert (b).
The scanpath begins at the black dot on the end of the dark blue segment. The first fixation is located at the black dot on the right and the color scheme shows the expert's fixation sequence as it progresses through time, going from dark blue to yellow to dark red. This expert demonstrates a typical scanpath that concentrates on the core and delta regions of the fingerprint, diagnostic of level one analysis. In contrast, the novice exhibits less directed behavior.
Fig 4.
An example of the attentional highlighting used in Experiments 1–3.
Each fixation location in a selected sequence shows attentional highlighting using a flashing red Gaussian intensity bump.
Fig 5.
The four phases of Experiments 1–3.
During the pre-test and post-test phases, subjects view images for 10 s and their fixations are recorded as they examine the prints. During the training phase, attentional highlighting guides fixations to locations of interest (red dot). Once subjects saccade to the location of the cue, the next fixation in the sequence is cued. Each trial begins with a fixation cue for 1 s preceding the onset of the fingerprint image. The fingerprint pairs shaded in purple represent the trained stimulus set, while the cyan fingerprint pairs represent the transfer stimulus set.
Fig 6.
Summary of spatial proximity results from Experiments 1–3.
Each experiment is shown in a row of the figure. The left and right columns present results for trained and transfer stimuli, respectively. Each graph shows the empirical cumulative distribution function before (blue) and after (red) the training phase. Each sample in the cumulative distribution function derives from the mean distance to all expert fixations (on the same image and impression) for a given subject fixation. Experiment 1–3 involved highlighted training based on expert fixations, novice fixations, and incongruent expert fixations, respectively.
Fig 7.
Summary of model-based analysis from Experiments 1–3.
Each experiment is shown in a row of the figure, and the left column presents results for trained stimuli and the right column for transfer stimuli. The ordinate of each graph indicates the log likelihood ratio (LLR), a measure of relative expertise. An increase in the LLR from before to after training indicates acquisition of expertise. Experiment 1–3 involved highlighted training based on expert fixations, novice fixations, and incongruent expert fixations, respectively. The error bars are calculated to account for systematic tendencies of data derived from the same subject and variability between subjects [43].
Table 1.
Low-level statistics of subject gaze behavior for Experiments 1, 2 and 3.
For each experiment, an analysis was performed on the before- and after-training data for the following statistics: the percentage of fixations that occur on the left impression versus the right (Left Fixations), the percentage of saccades that occur within an impression versus across impressions (Within Saccades), the mean fixation duration (Fixation Duration), the amplitude of saccades that occur on the left and right impression (Saccade Amplitude). Experiment 1–3 involved highlighted training based on expert fixations, novice fixations, and incongruent expert fixations, respectively.