Download presentation
Presentation is loading. Please wait.
Published byTheodore Black Modified over 8 years ago
1
UNCLASSIFIED Defense Forensic Science Center Development and Evaluation of a Model to Quantify the Weight of Fingerprint Evidence *Henry Swofford; Koertner A.J.; Salyards M.J. *Chief, Latent Print Branch, US Army Criminal Investigation Laboratory SAMSI Transition Workshop - 2016
2
UNCLASSIFIED Disclaimer The opinions or assertions contained herein are the private views of the author and are not to be construed as official or as reflecting the views of the Department of the Army or the Department of Defense. Names of commercial manufacturers or products included are incidental only, and inclusion does not imply endorsement by the authors, DFSC, OPMG, DA or DoD. Unless otherwise noted, all figures, diagrams, media, and other materials used in this presentation are created by the respective author(s) and contributor(s) of the presentation and research. 2
3
UNCLASSIFIED Fingerprint Examination (ACE) Analyze Impression Is Q & Q sufficient ? Compare with Reference End – Inconclusive Is Similarity sufficient ? End - “Identification” Yes No Yes
4
UNCLASSIFIED The problem... What is “Sufficient”? Sufficient Quality / Quantity of ridge information Sufficient Agreement of correspondence between two impressions
5
UNCLASSIFIED Defining “sufficient” Q & Q Three objectives: (1) Develop a tool capable of measuring the quality and quantity of fingerprint ridge information (2) Evaluate the quality metric against examiner performance metrics (3) Define “sufficient” quality and quantity of fingerprint ridge information as a function of the measured quality metric
6
UNCLASSIFIED DFIQI Defense Fingerprint Image Quality Index (DFIQI) Analysis Module
7
UNCLASSIFIED Image is segmented and thresholded “signal” “background” Dynamic mean thresholding (15 pixel [0.38mm] radius) Negative Image of fingerprint 100 pixels 2.54mm
8
UNCLASSIFIED Sharpness Ridge frequency Five variables measure clarity Contrast LQSr Ridge width S3PG
9
UNCLASSIFIED S3PG, Ridge Width; c/mm S3PG = % Pixels corresponding to “ridges” (~50%) Mean Ridge Width Spatial frequency of ridges (~2 ridges/mm)
10
UNCLASSIFIED (Contrast) Bimodal Separation “Signal” “Background”
11
UNCLASSIFIED Acutance (“Sharpness”) Acutance – Measures the average acutance for the entire ROI Acutance measures the physical characteristics that underlie the subjective perception of “sharpness” in an image – calculation adopted from Choong et al. (2003) *Choong et al. “Acutance, an Objective Measure of Retinal Nerve Fibre Image Clarity” Br. J. Ophthalmol 2003;87:322-6. p -2 x p -2 pixels p x p pixels 3 x 3 pixels I2I2 I3I3 I1I1 IcIc I4I4 I8I8 I6I6 I5I5 I7I7
12
UNCLASSIFIED Normalized local quality score evaluated using operationally developed “Good” and “Bad” fingerprint regions (images above are full fingerprints – regions of interest are 2.54mm x 2.54mm regions) 866 “Good” quality fingerprint ROIs and 3,699 “Bad” quality fingerprint ROIs Is LQSr any good at separating “good” vs. “bad”?
13
UNCLASSIFIED LQSraw “Good” ROIs “Bad” ROIs
14
UNCLASSIFIED END Analysis Module
15
UNCLASSIFIED The problem... What is “Sufficient”? Sufficient Quality / Quantity of ridge information Sufficient Agreement of correspondence between two impressions
16
UNCLASSIFIED Four objectives: (1) Develop a tool capable of measuring correspondence of fingerprint features (2) Evaluate how dissimilar prints can be when made from the same source (3) Evaluate how similar prints can be when made from different sources (4) Quantify the value of fingerprint correspondence in relation to the range of expected results under the conditions of same source and different source impressions Defining “sufficient”
17
UNCLASSIFIED DFIQI Defense Fingerprint Image Quality Index (DFIQI) Evaluation Module
18
UNCLASSIFIED 18 DFIQI – Conceptual Overview Image license free for public share and use from Pixaby.com
19
UNCLASSIFIED 19 DFIQI – Conceptual Overview
20
UNCLASSIFIED 20 DFIQI – Conceptual Overview
21
UNCLASSIFIED Quantifying Acceptable Variation Intra-Source Variability a.Repeated impressions from the same source collected on live- scan device at 500ppi: Ten different individuals (right thumb) One “flat” impression (non-distorted) Ten different “distorted” impressions Fifteen corresponding features annotated by latent print examiners Distorted image compared against non-distorted “flat” image
22
UNCLASSIFIED Quantifying Acceptable Variation Intra-Source Variability (Feature Movement) Raw distance movement n = 1,500 fingerprint minutiae
23
UNCLASSIFIED Quantifying Acceptable Variation Intra-Source Variability (Feature Movement) Sqrt (Raw Distance Movement) n = 1,500 fingerprint minutiae
24
UNCLASSIFIED Quantifying Acceptable Variation Intra-Source Variability (Angle Movement) Raw angle difference (radians) n = 1,500 fingerprint minutiae
25
UNCLASSIFIED Intra-Source variability Intra-Source Variability a.Repeated impressions from the same source collected on live- scan device at 500ppi: Fifty different individuals (right thumb) One “flat” impression (non-distorted) Ten different “distorted” impressions Fifteen corresponding features annotated by latent print examiners Distorted image compared against non-distorted “flat” image
26
UNCLASSIFIED Compare against several distorted prints from the same source Image license free for public share and use from Pixaby.com Intra-Source variability
27
UNCLASSIFIED Inter-Source variability Inter-Source Variability a.200 Fingerprint images a.100 representing the “core” region of a fingerprint b.100 representing the “delta” region of a fingerprint b.n fingerprint features were annotated on each impression, cropped down to only a partial impression represented by the annotated features, and searched through a operational AFIS database (database contains ~12,000,000 individuals; ~100,000,000 different fingerprints) c.The #1 search candidate was exported (most similar candidate according to AFIS algorithms). New search for each quantity of features (n range from 5 – 15 features) d.m fingerprint features were annotated on the candidate result – annotations performed indiscriminately across the core and delta, respectively (where m > n+5)
28
UNCLASSIFIED Doe #1 Doe #2 Doe #3 Doe #4 Doe #n... Compare against the most similar print from different sources (~100,000,000 prints) AFIS Image license free for public share and use from Pixaby.com Inter-Source variability
29
UNCLASSIFIED Intra- & Inter-Source variability Intra & Inter-Source Variability (Delta) DFIQI Score # Minutiae
30
UNCLASSIFIED Intra- & Inter-Source variability Intra & Inter-Source Variability (Delta) Sensitivity 1 - Specificity
31
UNCLASSIFIED Intra- & Inter-Source variability Intra & Inter-Source Variability (Core) DFIQI Score # Minutiae
32
UNCLASSIFIED Intra- & Inter-Source variability Intra & Inter-Source Variability (Core) Sensitivity 1 - Specificity
33
UNCLASSIFIED Optimizations and Challenges 1.Improving global scoring methods 2.Increased datasets for “different source” prints 3.Evaluation against latent impressions left under uncontrolled conditions 4.Evaluate relationship between quantitative scoring and examiner performance during comparison exercises
34
UNCLASSIFIED Next Steps Continue working with relevant statisticians, academia, and legal stakeholders to identify the appropriate scope of use and optimization Transition technology into public domain for use by other federal, state, local agencies with the intent to help advance the discipline at large Integrate into casework operations as a means of quantifying, defining, and standardizing practice
35
UNCLASSIFIED Henry Swofford, MSFS Chief, Latent Print Branch USACIL 35 Henry.J.Swofford.civ@mail.mil (404) 469-5611
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.