LatentSleuth is a novel software toolset designed for (and with direct feedback from) latent print examiners. The toolset includes a small-scale automated fingerprint identification system (AFIS) matcher that leverages a novel matching algorithm that calculates a warp between a latent print image and a given reference print image.
The resultant warp allows any point in “latent space” to be mapped to “reference space” (and vice versa). The non-linear nature of the warp allows the matching algorithm to fully account for the elasticity of skin that creates the significant distortion differences between latent prints and associated reference prints (a characteristic that makes latent print matching difficult by nature).
The ‘small-scale’ (but powerful) nature of LatentSleuth’s underlying matcher algorithm makes it ideal for scenarios where the number of identities being compared is low, such as matching against an expanded return set from a large scale AFIS matcher, or running against tenprints from suspected individuals or identities that are not in AFIS databases.
To determine the accuracy and validity of the LatentSleuth software for use in casework, 600 automated searches were completed using 200 latent prints with true-mated reference prints. Fifty latent prints of four different quality levels (high, medium high, medium low, and low) were searched against groups consisting of three, five, or ten sets of reference images (tenprint cards). Medium low and low quality latent prints were searched with automated processing edits and again with manual processing, while high and medium high latent prints were only searched with automated processing.
The LatentSleuth software provided accurate results in all latent print quality levels against all three levels of comparison complexity and was deemed suitable for use in casework. The technology was then implemented into casework to determine if it improved the efficiency of the comparison workflow and/or the accuracy of results.
The search phase of traditional manual comparisons was compared against the time spent using the software: uploading images and reviewing recognition results. One hundred and thirteen cases were completed by two examiners, one examiner conducing manual comparisons, one examiner using LatentSleuth.