Emerging Fingerprint Methods: 3D Capture, Age Estimation, ML

A conventional flat-bed optical fingerprint scanner captures a 2D projection of the contact surface of a finger pressed flat against a glass platen. The contact area is typically the central pad of the fingertip. Ridge detail at the sides of the finger, the fingertip arch, and the area near the nail folds is not captured. For biometric enrolment, this is an acceptable loss: the central pad contains the core pattern area and the most reliable ridge detail for AFIS matching. For forensic casework, the limitation is more significant: a latent print deposited when a finger grips a cylindrical object such as a bottle, a gun barrel, or a door handle may contain ridge detail primarily from the sides of the finger, which a flat 2D scan cannot fully resolve.

The TBS (Touchless Biometric Systems AG, a Swiss biometrics manufacturer) 3D Enroll terminal represents one of the most widely deployed implementations of contactless 3D fingerprint capture for access control and border enrolment. The system uses structured-light projection, in which a grid of light is projected onto the finger and the distortion of the grid reveals the 3D surface profile. The resulting point cloud is converted to a 3D ridge-detail map, which can be matched against flat-enrolled templates by re-projecting the 3D surface into a flat plane at multiple angles and selecting the best-matching projection. Thales reports that 3D capture significantly reduces the failure-to-enrol rate in populations where flat-platen scanning produces poor quality images, including:

• Elderly individuals (ridge detail attenuated by age-related skin changes).
• Manual workers (worn or scarred ridges).
• Individuals with hyperhidrotic or dry skin conditions.

By 2023, Thales 3D Enroll terminals were operational at border crossing points in France, the Netherlands, the United Arab Emirates, and Australia, among others.

In the laboratory, the challenge is different. A latent print developed on a curved surface (a wine bottle, a glass pipe, a car steering wheel) is typically photographed with conventional 2D digital macro-photography. The curvature of the surface distorts the photograph: ridge detail at the edges of the print is foreshortened relative to the centre, and the focal plane can only be optimal across the print's full area at the expense of increased depth of field, which adds noise. Photogrammetric methods can partially address this but require the surface and print to be stationary during imaging.

Micro-computed tomography (micro-CT) represents the highest-resolution approach to 3D fingerprint recovery from laboratory exhibits. A group at University College London demonstrated in 2019 that latent prints developed with aluminium powder on polymer surfaces could be recovered by micro-CT scanning at resolutions sufficient to reveal individual ridge units. The 3D nature of the scan could distinguish between a print on the surface and a smear or substrate artefact embedded in the surface material. The technique is non-destructive and preserves the exhibit for further examination, but micro-CT scanner time is expensive and the method has not yet been operationally adopted by UKAS-accredited UK forensic providers.

The forensic and border-crossing use cases for 3D fingerprint capture are analytically related but operationally distinct. At a border, the objectives are:

1. High throughput (thousands of travellers per hour at a major port of entry).
2. Low failure-to-enrol rates.
3. Compatibility with national biometric databases typically built from flat-scanned enrolments.

The third objective is the most technically complex. Most national biometric passport systems and border-crossing databases hold fingerprint templates derived from flat-platen scanning, compliant with the ISO/IEC 19794-4 standard for fingerprint image data and ISO/IEC 29109-4 for conformance testing. A 3D capture system that re-projects to produce a flat image in the standard format can match against these legacy templates, but the re-projection introduces choices (which angle produces the best match to the enrolled template?) that affect match accuracy. The NIST Fingerprint Vendor Technology Evaluation (FpVTE) series, most recently the NIST MINEX III evaluation for minutia interoperability, has begun to include 3D-derived template evaluations alongside conventional flat-scan templates. Full evaluation datasets and published accuracy benchmarks for cross-dimensional matching were not yet available as of 2024.

Interpol's AFIS currently operates on flat 2D fingerprint images compliant with the INTERPOL Best Practice Guidelines. The migration path for 3D templates into Interpol AFIS is an active standardisation topic within the Technical Working Group on Fingerprints. India's National Automated Fingerprint Identification System (NAFIS), operated by the National Crime Records Bureau (NCRB), holds enrolment templates in flat 2D format under NIST standards, and no 3D enrolment programme was active at NCRB by 2025.

The interval between a fingerprint's deposition and its recovery is directly relevant to the weight of the evidence: a fingerprint on a murder weapon deposited within 24 hours of the offence carries different significance from one deposited three weeks earlier. Until recently, no validated method existed for estimating deposition time.

Latent fingerprint residue is a complex mixture of sebaceous secretions (lipids including wax esters, squalene, and fatty acids), eccrine secretions (amino acids, salts, glucose, lactate, and other water-soluble compounds), and environmental contamination. The same chemical composition is explored in the context of cyanoacrylate fuming and fluorescent dye-stain development and in chemical methods for porous surfaces. The chemical composition of the residue changes over time due to oxidation, volatilisation, microbial degradation, and photodegradation. The research programme on age estimation attempts to identify chemical changes reliable enough to serve as a molecular clock.

The most developed line of research focuses on amino acid racemisation. Eccrine sweat contains free amino acids including serine, glycine, and aspartic acid. These amino acids undergo racemisation over time: the biologically produced L-form converts gradually to the D-form at a rate that depends on temperature, humidity, and the substrate. Research from the Aalto University forensic chemistry group (Finland) and from the Forensic Science International research cluster at the University of Technology Sydney has characterised racemisation rates for aspartic acid in fingerprint residue on glass and polymer surfaces. The D/L ratio of aspartic acid increases measurably over periods of 1 to 30 days under controlled conditions. A fingerprint deposited 24 hours ago shows a significantly lower D/L ratio than one deposited 14 days ago on the same surface under the same temperature regime.

The lipid oxidation approach focuses on the degradation of squalene, a polyunsaturated hydrocarbon abundant in sebaceous secretions that is highly susceptible to oxidation. Research from the ENFSI Working Group on Fingerprints, particularly contributions from the Netherlands Forensic Institute and the LGC (Laboratory of the Government Chemist, UK), has shown that squalene oxidation products (squalene monoepoxide, squalene diepoxide, and further oxidation products) accumulate over time in a roughly predictable pattern. Mass spectrometry imaging, using matrix-assisted laser desorption ionisation mass spectrometry (MALDI-MSI) or desorption electrospray ionisation (DESI-MSI), can image the spatial distribution of squalene and its oxidation products across the fingerprint in situ, without chemically developing the print.

Mass spectrometry imaging has transformed the analytical side of fingerprint age estimation because it can map the chemical composition of a fingerprint spatially without the need to dissolve or extract the print. MALDI-MSI fires a laser at points across the fingerprint surface in a grid pattern, ablating and ionising the residue at each point, and records a full mass spectrum at each pixel. The resulting dataset is a three-dimensional array (x position, y position, mass-to-charge ratio) from which maps of any compound's distribution can be extracted. For age estimation, the relevant maps are the ratio of squalene to its oxidation products across the print area.

The DESI-MSI approach, developed by the Cooks group at Purdue University and adapted for fingerprint analysis by groups at the University of Surrey and at University of Technology Sydney, uses a charged solvent spray to desorb surface molecules without laser ablation. It is gentler on the substrate and potentially applicable to a wider range of surface types. Both MALDI and DESI imaging can be performed without chemically developing the print. This is significant because development permanently alters the residue composition and would compromise any subsequent age-estimation analysis. The workflow requirement is therefore that age estimation must occur before any chemical or powder development step, which has implications for how forensic laboratories sequence their examination of latent fingerprints.

Open research questions as of 2025 include:

• Calibration of racemisation and oxidation rates across the full range of environmentally realistic temperature and humidity profiles.
• The effect of substrate type (glass, plastic, metal, paper, textiles) on degradation rates.
• The effect of individual donor variation in secretion composition on baseline ratios.
• The statistical framework for translating a measured D/L ratio or oxidation-product ratio into an estimated age with confidence intervals suitable for evidentiary use.

The European Association of Forensic Science (EAFS) and ENFSI Working Group on Fingerprints both listed fingerprint age estimation as a priority research topic for the 2023-2027 planning period.

In India, the CFSL network and state FSLs do not yet have operational MALDI-MSI or DESI-MSI capabilities. Access to research-grade mass spectrometry imaging infrastructure is concentrated in a small number of academic institutions (IIT Delhi, CSIR-NEERI Nagpur, and a handful of NIPERs). Fingerprint age estimation, even when validated for court use, will initially be available only through expert witness testimony from academic institutions rather than through accredited forensic service providers. Similar gaps exist in many lower-middle-income country forensic laboratory systems globally.

The ACE-V methodology, formalised by the International Association for Identification (IAI) and adopted by the Scientific Working Group for Friction Ridge Analysis, Study and Technology (SWGFAST), structures fingerprint comparison into four steps: Analysis of the latent print in isolation, Comparison with the reference print, Evaluation of the comparison findings, and Verification by a second independent examiner. ACE-V is not an algorithm; it is a procedural framework that specifies the order of operations and documentation requirements but leaves the substantive analytical decisions to the examiner.

Machine learning tools are being introduced at each step of this workflow. At the Analysis stage, automated quality scoring algorithms estimate the number of usable friction ridge units in a latent print and assign a quality grade, helping the examiner decide whether a print is suitable for comparison or should be excluded as insufficient for individualization. The NIST Latent Print Quality Metric, developed by NIST, uses a convolutional neural network trained on expert-annotated quality ratings to produce a quality score compatible with the ANSI/NIST ITL fingerprint standard.

At the Comparison stage, automated similarity-ranking tools present the examiner with a ranked list of candidate matches from the AFIS database, ordered by a computed similarity score. This is not new in principle: AFIS systems have done this since the 1970s. What is new is the adoption of deep learning similarity metrics that substantially outperform classical minutia-matching algorithms on degraded and partial prints. NIST FpVTE 2019 evaluated 195 algorithms from 48 providers. The best-performing system achieved a rank-1 identification rate of 99.67 per cent on the NIST SD301 dataset, compared to approximately 97 per cent for the best classical minutia-based system in FpVTE 2012. These improvements are particularly pronounced for partial prints at below 50 per cent contact area, which constitutes the majority of casework latent prints.

At the Evaluation stage, emerging tools present the examiner with a predicted minutia map overlaid on the latent print image, automatically annotating ridge endings, bifurcations, and dots as a starting point for the examiner's own markup. Research from the FBI Laboratory and from the Henry C. Lee Institute of Forensic Science has shown that automated minutia extraction tools reduce inter-examiner variability in marking: examiners presented with an automated markup tended to agree more with each other and with the automated tool than examiners marking from scratch. Whether this represents a genuine reduction in error or a homogenisation of examiner judgment is an active discussion in the forensic community. The cognitive bias implications of shared tool outputs are discussed in the Dror 2006 topic.

Fingerprint evidence is among the most widely admitted forms of forensic science evidence in courts worldwide, but its admissibility was built on a century of practice rather than on an independently validated error-rate database. The 2009 National Academy of Sciences report "Strengthening Forensic Science in the United States" identified fingerprint examination as lacking rigorous, population-level error-rate data and challenged the field to develop such data. The PCAST (President's Council of Advisors on Science and Technology) 2016 report similarly concluded that the foundational validity of latent fingerprint analysis had been established but that the field lacked adequate error-rate estimates for partial prints at realistic casework quality levels.

The US Department of Justice's Uniform Language for Testimony and Reports (ULTR) standards for fingerprints, finalised in 2021, were a direct response to PCAST. They restrict fingerprint examiners to saying that they identified, excluded, or were unable to determine, with standardised explanatory language. Claims of certainty ("with 100 per cent certainty" or "to the exclusion of all other persons") not supported by population-level error-rate data are prohibited.

In England and Wales, fingerprint evidence is admitted under the framework established in R v. Turner [1975] and developed through subsequent Court of Appeal decisions. The Attorney General's Guidelines on Disclosure (2022) require full disclosure of any quality issues with fingerprint evidence, including any software tools used in analysis. The UK Forensic Science Regulator's Codes of Practice and Conduct include fingerprint analysis in their scope, requiring UKAS accreditation for evidential casework. The Cognitive Bias Study Group within the IAI Friction Ridge Division has published guidelines on blind verification and on the sequencing of information presented to examiners to reduce context-induced bias, addressing a finding from the Dror and Charlton (2006) study that UK fingerprint examiners could be induced to reverse prior identifications when presented with contextual information suggesting innocence.

1. Latent print recovery
   Photograph the exhibit in the condition it was received. If age estimation is contemplated, proceed to chemical-free analysis (MALDI-MSI or DESI-MSI) before any powder or chemical development. If age estimation is not required, proceed to standard development workflow.
2. Quality assessment (enhanced by ML)
   Apply automated quality scoring (NIST Latent Quality Metric or equivalent) to the developed print image. Record the quality score in the case notes. Use the score to inform, not replace, the examiner's sufficiency decision.
3. AFIS search
   Submit the latent print image to the AFIS (with deep-learning similarity ranking where available). Review the candidate list. Document the AFIS system version, the database searched, the number of candidates reviewed, and the rank position of any candidate selected for ACE-V comparison.
4. ACE-V comparison
   Conduct Analysis (in isolation), Comparison (against the AFIS candidate), Evaluation (express a conclusion as per ULTR or FSR guidance). Use automated minutia extraction as a starting-point tool only, reviewing all automated markings against the image before acceptance.
5. Verification
   Second independent examiner repeats ACE steps without knowledge of the primary examiner's conclusion (blind verification per IAI Cognitive Bias guidelines and FSR Codes of Practice). Blind verification is operationally mandated in FBI casework and recommended in UK casework from 2022.
6. Reporting
   Express conclusions in standardised ULTR language (US) or FSR-compliant language (UK), citing the specific tools used, their versions, and any quality limitations of the latent print. Disclose any algorithmic tools in the case report as required by Attorney General disclosure guidance (UK) or Federal Rule of Evidence 702 obligations (US).