international joint conference on ieee first latent-to
TRANSCRIPT
Latent-to-full Palmprint Comparison Based on
Radial Triangulation under Forensic Conditions Ruifang Wang, Daniel Ramos and Julian Fierrez
[email protected], [email protected], [email protected]
Biometric Recognition Group - ATVS Escuela Politecnica Superior
Universidad Autonoma de Madrid
http://atvs.ii.uam.es
In this paper, we develop latent-to-full palmprint identification
technology based on radial triangulation as a first step towards a
quantification of the value of evidence in palmprints using
likelihood ratios. Firstly, palmprint preprocessing and post-
processing are implemented for full prints feature extraction by a
commercial biometric SDK MegaMatcher 4.0 in an automatic way,
while features of latent prints are manually extracted by forensic
experts. Then a latent-to-full palmprint comparison algorithm
based on radial triangulation is proposed. 22 latent palmprints
from real forensic cases and 8680 full palmprints from criminal
investigation field are used for performance evaluation.
Experimental results proof the usability and efficiency of the
proposed system, i.e., rank-1 identification rate of 62% is
achieved despite the inherent difficulty of latent-to-full palmprint
comparison. This shows the adequacy of radial triangulation as a
useful way of extracting information on identity from palmprints.
ABSTRACT MOTIVATION
EXPERIMENTAL RESULTS
LATENT-TO-FULL PALMPRINT COMPARISON
S.K. Dewan et al. (The New York Times 2003): Surveys of law enforcement agencies which indicate that 30 percent of the latents recovered from crime scenes are from palms.
Jain et al. (IEEE TPAMI 2009): Rank-1 identification rate 69% is far from the needs in forensic applications.
Ian Evett (Science and Justice 2011): It is necessary for the scientist to consider the probability of the observations given each of the stated propositions, i.e., quantification of evidential value.
C. Neumann and C. Champod (JFS 2007): Radial triangulation was proposed as a step towards quantification of evidential value in fingerprints successfully.
• The usability and efficiency of the proposed latent-to-full palmprint identification system based on radial triangulation have been proved by experimental results.
• Radial triangulation has the adequacy to serve as a method of local minutiae structure modeling used for minutiae-based palmprint comparison.
• The existence of a large number of spurious minutiae significantly deteriorates the performance of latent-to-full palmprint comparison, while image preprocessing and post-processing improve the performance.
• The centroid of the polygon generated by radial triangulation can be used to select calibration center for the alignment of latent and full palmprint minutiae sets to improve the accuracy of global comparison.
• Developing a model based on radial triangulation for the computation of likelihood ratios using the evidence in palmprints can be feasible.
CONCLUSIONS SELECTED REFERENCES S.K. Dewan, “Elementary, Watson: Scan a Palm, Find a Clue,” The New York Times, Nov. 2003, http://www.nytimes.com/. A.K. Jain and J. Feng. Latent Palmprint Matching. IEEE Transactions on Pattern Analysis and Machine Intelligence, 31(6):1032-1047, 2009. C. Neumann, C. Champod, R. Puch-Solis, N. Egli, A. Anthonioz and A. Bromage-Griffiths. Computation of Likelihood Ratios in Fingerprint Identification for Configurations of Any Number of Minutiae. Journal of Forensic Science, 52(1):54-64, 2007. Expressing Evaluative, Opinions:a position statement. Science & Justice, 51(2011):1-2. Y. Zheng, G.S. Shi, L. Zhang, Q.R. Wang and Y.J. Zhao. Research on Offline Palmprint Image Enhancement. IEEE International Conference on Image Processing (ICIP’07), 2007, pp. 541-544.
Our goal is to develop forensic automatic palmprint identification system based on radial triangulation as a first step towards a quantification of the value of evidence in palmprints.
International Joint Conference on
Biometrics
IEEE First
Latent minutiae set Full minutiae set
Radial triangulation Radial triangulation
Local minutiae comparison
Alignment
Global minutiae comparison
Comparison score
computation
Average minutiae
number
Image
number
Latent palmprints 43 22
Full
palmprints
Original
(RT1) 2191
8680 With preprocessing
(RT2) 1268
With pre- and post-processing
(RT3) 1023
N=9
Database Comparison results
Latent Full Rank-1 Rank-20 Average
time(ms)
RT1 22 8680 41% 61% 230
RT2 22 8680 55% 68% 140
RT3 22 8680 62% 70% 100
MinutiaCode 100 10200 69% 76% 340
DATABASES
Manually Automatically
RADIAL TRIANGULATION
PREPROCESSING
POST-PROCESSING
SCORE COMPUTATION
Segmentation Orientation
field
Thinned
image
N
kkkkkkC STLRVPLS 11, },,,{,[
ji
ji
lk
N
l
N
k
lk
jimDV
DVN
lks,
1 1
,2
1
),(
Local structure
comparison
Minutiae similarity
qn SSS 20
M
Mn
N
NS and
FM
M
LM
MDq
NN
N
NN
NSS
minutiae filter based on
line feature extraction
Latent palmprint database:
22 images from forensic cases
Full palmprint database:
8680 images
4340 subjects x 2 images
Image size: 2304×2304