Enhanced Latent Fingerprint Segmentation through Dictionary Based Approach

Scientific Journal Impact Factor (SJIF): 1.711
International Journal of Modern Trends in Engineering
and Research
www.ijmter.com
@IJMTER-2014, All rights Reserved 121
e-ISSN: 2349-9745
p-ISSN: 2393-8161
Enhanced Latent Fingerprint Segmentation through Dictionary Based
Approach
Nithya.K1
, Karthi Prem.S2
and Prem.P.E3
1
Department of IT, Vivekanandha College of Engineering for Women
2
3
Abstract— The accuracy of latent finger print matching compared to roll and plain finger print
matching is significantly lower due to background noise, poor ridge quality and overlapping
structured noise in latent images. In this paper the proposed algorithm is dictionary-based approach
for automatic segmentation and enhancement towards the goal of achieving “lights out” latent
identifications system. Total variation decomposition model with L1 fidelity regularization in latent
finger print image remove background noise. A coarse to fine strategy is used to improve robustness
and accuracy. It improves the computational efficiency of the algorithm.
Keywords- Latent finger print, image decomposition, ROI, segmentation, ridge enhancement.
I. INTRODUCTION
Latent fingerprints refer to fingerprints got from the surfaces of objects acknowledgeable touched
or handled by a person typically at crime scenes. Compared to rolled and plain finger prints, latent
finger print is typically poor in quality and complex background noise. Due to this problem latent
finger print accuracy is much lower.
The main challenging problem in latent identification system is how to automatically extract
reliable features in latent, especially latent with poor quality. The various features are proposed like
ROI (Region of Interest), singular points and minutiae detection.
The human factor issue in latent examinations has raised some reliability and repeatability.
Brandon Mayfield case is involved to identify mistakes based on latent matching. The lights out
capability develop a fully automatic latent feature extraction module. This is highly desirable to
(i) Increase the throughput of latent matching systems,
(ii) Improve repeatability of latent feature extraction and,
(iii) Increase the compatibility
Between features extracted in the latents and features extracted in reference images.

International Journal of Modern Trends in Engineering and Research (IJMTER)
Volume 01, Issue 05, [November - 2014] e-ISSN: 2349-9745, p-ISSN: 2393-8161
Figure.1Three types of fingerprint images. (a) Rolled fingerprint, (b) plain fingerprint, (c) latent fingerprint with
foreground (friction ridge pattern) highlighted by red outline. Notice the presence of different types of noise and
distortion in (c).
Commonly rolled/plain print matching or latent matching typically contains a number of
modules, including region of interest (ROI) Segmentation (separating friction ridge from
background), enhancement, feature extraction, and matching.
To address the two main difficulties in latent fingerprint matching, namely presence of
structured noise and poor quality of ridge structure, we propose a dictionary-based segmentation and
enhancement of latent fingerprint. To remove the structured noise, the total-variation model with L1
fidelity regularization is used to decompose a latent into a texture part and a cartoon part. The main
contributions of this paper are as follows:
1. Ridge structure dictionary is proposed for latent segmentation and enhancement. The
dictionary is learnt from high quality fingerprint patches.
2. The ridge quality of a patch is defined as the structural similarity between the patch and its
reconstruction from the learnt ridge structure dictionary. Orientation and frequency fields of a
patch are estimated from its reconstruction.
3. To balance robustness and accuracy, a coarse to fine strategy is proposed which uses
dictionaries at two levels of resolution (64 × 64 and 32× 32).
4. The proposed segmentation and enhancement algorithms outperform published algorithms
and can also significantly boost the performance of a state of the-art commercial latent
matcher on two latent databases (NIST SD27 and WVU DB).
II. RELATED WORK
Various researches have been reported in this related work based on segmentation and
enhancement of latent finger print [2]-[6]. The idea of “Latent Fingerprint Enhancement via Robust
Orientation Field Estimation” demonstrates the improvement of the automatic match accuracy and it
is effective which help to find a correct orientation for latent fingerprint algorithm. [1].Hence it
improved the performance of the orientation field.
The Short Time Fourier Transform (STFT) improves the evaluation of the enhancement
algorithm. The STFT analyses consider the improvement only in matching the accuracy of the poor
quality print [15]. It uses some set of data base images for analysis purposes, which shows 17% of
improvement in recognition rate.
The Adaptive Directional Total Variation (ADTV) is derived for the latent fingerprint
segmentation. It gives better accuracy in segmentation and enhancement process. Hence it increases
the performance of the latent fingerprint feature detection and matching.

For image segmentation and estimation of fingerprint by ridge orientation a systematic
method is used [2]. Since the ridge structure is taken by using the Gabor filter method, which
improves the matching performance. This systematic method automatically removes the broken
ridge. Thus it provides the accuracy in the minutiae detection.
Some of the systematic methods are used for computation of the directional field and singular
point of the fingerprint. By using the averaged square gradient and singular-point-extraction the error
rate of singular point is improved for the purposes of detecting the low quality fingerprint.
The K-mean Singular Value Decomposition (K-SVD) is used for adaptive dictionaries to
achieve the spares single representation [4]. The problem of the scalability in K-SVD is more
complement so it can be improved in future.
Latent fingerprint enhancement algorithm is used for finding the region of interest (ROI) and
Singular Point (SP) [1]. It improves the accuracy matching for fingerprint which obtains the multiple
orientation elements for each image block.
III. PROPOSED ALGORITHM
While current automated fingerprint identification systems have achieved high accuracy in
matching rolled/plain prints, latent matching still remains to be a challenging problem and requires
much human intervention. The goal of this work is to achieve accurate latent segmentation, which is
an essential step towards achieving automatic latent identification. Existing fingerprint segmentation
algorithms performs poorly on latent prints, as they are mostly based on the assumptions that are
only applicable for rolled/plain fingerprints.
However, the proposed algorithm still does not work well on very poor quality latent
fingerprint images. Latent fingerprints found at crime scenes provide crucial evidence to law
enforcement agencies. The latents are typically searched against a large fingerprint database which is
the collection of rolled/plain fingerprints of previously apprehended criminals. Due to the poor
quality of the latents, latent examiners perform manual feature markup and
Visual verification between the latent and the candidate fingerprints from the database.
The latent finger print segmentation and enhancement consists of the following steps:
(i) Latent decomposition
(ii) Coarse level estimation of quality map and orientation and frequency fields.
(iii) Fine level estimation of quality map and orientation and frequency fields.
(iv) Segmentation and enhancement.
A “Lights-Out” mode for latent identification is desired to reduce the burden on latent examiners
and to introduce a level of consistency in fingerprint matching, particularly in searching ever
growing fingerprint database.
The proposed module latent matching or search is still a challenging problem due to presence of
complex background noise and poor quality of friction ridge structure in many latents proposed an
automatic latent segmentation and enhancement algorithm based on image decomposition and coarse
to fine ridge structure dictionaries. The two different latent fingerprint databases, NIST SD27 and
WVU DB, in conjunction with three different COTS matchers show that the proposed algorithm is

able to improve the performance of two COTS ten print matchers and can even boost the
performance of a state-of-the-art latent matcher by weighted match score fusion. However, the
proposed algorithm still does not work well on very poor quality latent fingerprint images.
3.1. Algorithm Overview
The proposed algorithm consists of an off-line dictionary learning stage and on-line stage for
segmentation and enhancement. In the off-line stage, two types of dictionaries are learnt:
i) a coarse-level dictionary with patch size of 64 × 64 pixels which is used for coarse estimation of
ridge quality map, orientation and frequency fields, and
ii) 16 fine-level dictionaries with patch size of 32 ×32pixels which are used for fine estimation of
ridge quality map, and orientation and frequency fields computation;
Given the dictionaries, on-line latent segmentation and enhancement consists of the following
steps:
3.1.1. Latent decomposition:
Input latent is decomposed into cartoon and texture images via local total variations, the
cartoon image is discarded. The relative reduction rate before and after filtering the image with the
low-pass filter measures the local oscillatory behavior which is given by
ߣߪሺ‫ݔ‬ሻ =
௅்௏ఙሺ௙ሻሺ௫ሻି ௅்௏ఙሺ௅ఙ∗௙ሻሺ௫ሻ
௅்௏ఙሺ௙ሻሺ௫ሻ
where, Lσ is an σ-sized low-pass filter, x belongs to the cartoon part. The relative reduction rate is
ߣߪሺ‫ݔ‬ሻ
3.1.2. Coarse level estimation of quality map and orientation and frequency fields.
The texture image is divided into overlapping patches of size 64 × 64 pixels. The sparse
representation of p can be obtained by solving the following optimization problem using orthogonal
matching pursuit. The approximation can be calculated as,‫݌‬ෝ= ‫ܦ‬௦
௖
α , where α is given by
α = ሺ‫ܦ‬௦
௖்
‫ܦ‬௦
௖
ሻିଵ
‫ܦ‬௦
௖்
‫݌‬. The residual vector can be calculated as r(p) = p - ‫̂݌‬ = p - ‫ܦ‬௦
௖்
α'
3.1.3. Fine level estimation of quality map and orientation and frequency fields.
The texture image is divided into overlapping patches of size 32 × 32 pixels. Each patch has
32 ×16 or 16 ×32 overlapping pixels with each of its four connected neighboring blocks. For each
patch, the coarse ridge orientation value is first used to index the corresponding fine-level dictionary.
The fine estimation of ridge quality map, orientation and frequency fields are computed as
coarse estimations.
3.1.4. Segmentation and enhancement
The final quality map Q is computed as the average of coarse level quality and fine level
quality by,

Q = ½(ܳ஼
+ܳ௙
)
The quality map Q is then normalized to the range [0,1]. The orientation and frequency
parameters of the filter are tuned based on the fine-level orientation field (uf ) and the average
frequency of coarse-level frequency field and fine level frequency field (ffþfc)the standard deviation
of the Gaussian envelope in the Gabor filter The coarse quality map and fine quality map are
combined for latent segmentation. In the foreground of texture image, Gabor filtering based on the
orientation and frequency fields applied for latent enhancement.
V. ARCHITECTURAL DIAGRAM
The analysis of Image decomposition ROI segmentation and enhancement using coarse to
fine strategy to increase accuracy of latent finger print image. The overview of proposed latent
segmentation and enhancement architecture is been shown in fig.2
The latent finger print image is been acquired. Then image decomposition method is used to get
the output as textured image. Coarse to fine strategy is used to estimate texture image.
Dictionary construction method using this estimation consists of two types 1. coarse and 2. fine
strategy.
Initially the off-line coarse dictionary construction is used and the finger print image is acquired.
The path extraction method is applied get training samples of image the finally dictionary learning
method is applied to get the outcomes of dictionary elements. The data are been sent to coarse level
and quality map, orientation and frequency field are estimated in it.
Figure.2 Proposed Latent Segmentation and enhancement architecture
Then apply the same process for the off-line fine dictionary. The off-line fine dictionary
construction is used and the finger print image is acquired. The path extraction method is applied get
training samples of image the finally dictionary learning method is applied to get the outcomes of
dictionary elements. The data are been sent to coarse level and quality map, orientation and
frequency field are estimated in it.
The result of latent segmentation and enhancement are derived by combining the coarse
estimation and fine estimation which are represented as ROI segmentation.

VI. CONCLUSION
The new method has been proposed to improve the computational efficiency of the algorithm.
A coarse to fine strategy is used to improve robustness and accuracy. The latent finger print
matching compared to roll and plain finger print matching, accuracy is significantly lower due to
background noise, poor ridge quality and overlapping structured noise in latent images. In this paper
the proposed algorithm is dictionary-based approach for automatic segmentation and enhancement
towards the goal of achieving “lights out” latent identifications system. Obtaining the ROI
segmentation and enhancement through coarse to fine ridge strategy.
REFERENCES
[1] S. Yoon, J. Feng, and A. Jain, “Latent Fingerprint Enhancement via Robust Orientation Field Estimation,” Proc. Int’l
Joint Conf. Biometrics (IJCB), pp. 1-8, 2011.
[2] E. Zhu, J. Yin, C.Hu, and G. Zhang, “A Systematic Method for Fingerprint Ridge Orientation Estimation and Image
Segmentation,” Pattern Recognition, vol. 39, no. 8, pp. 1452-1472, 2006.
[3] J. Feng, J. Zhou, and A.K. Jain, “Orientation Field Estimation for Latent Fingerprint Enhancement,” IEEE Trans.
Pattern Analysis and Machine Intelligence, vol. 35, no. 4, pp. 925-940, Apr. 2013.
[4] M.Aharon, M.Elad, and A. Bruckstein, “K-SVD: An Algorithm for Designing Overcomplete Dictionaries for Sparse
Representation,” IEEE Trans. Signal Processing, vol. 54, no. 11, pp. 4311- 4322, Nov. 2006.
[5] J. Zhang, R.Lai, and C.-C. Kuo, “Adaptive Directional Total-Variation Model for Latent Fingerprint Segmentation,”
IEEE Trans. Information Forensics and Security, vol. 8, no. 8, pp. 1261-1273, Aug. 2013.
[6] B.T. Ulery, R.A. Hicklin, J. Buscaglia, and M.A. Roberts, “Accuracy and Reliability of Forensic Latent Fingerprint
Decisions,” Proc. Nat’l Academy of Sciences of USA, vol. 108, no. 19, pp. 7733-7738, 2011.
[7] C. Wilson, R.A. Hicklin, H. Korves, B. Ulery, M. Zoepfl, M. Bone, P. Grother, R. Micheals, S. Otto, and C. Watson,
“Fingerprint Vendor Technology Evaluation 2003: Summary of Results and Analysis Report,” Report NISTIR 7123,
Nat’l Inst. of Standards and Technology, 2004.
[8] M.D. Indovina, V. Dvornychenko, R.A. Hicklin, and G.I. Kiebuzinski, “Evaluation of Latent Fingerprint
Technologies: Extended Feature Sets (Evaluation 2),” Technical Report NISTIR7859, Nat’l Inst. of Standards and
Technology, 2012.
[9] B.T. Ulery, R.A. Hicklin, J. Buscaglia, and M.A. Roberts, “Repeatability and Reproducibility of Decisions by Latent
Fingerprint Examiners,” PloS one, vol. 7, no. 3, article e32800, 2012.
[10] N.J. Short, M.S. Hsiao, A.L. Abbott, and E.A. Fox, “Latent Fingerprint Segmentation Using Ridge Template
Correlation,” Proc. Fourth Int’l 2008 Conf. Imaging for Crime Detection and Prevention, pp. 1- 6, 2011.
[11] H.-S. Choi, M. Boaventura, I.A.G. Boaventura, and A.K. Jain, “Automatic Segmentation of Latent Fingerprints,”
Proc. IEEE Biometrics: Theory, Applications and Systems (BTAS), 2012.
[12] X.J. Chen, J. Tian, J.G. Cheng, and X. Yang, “Segmentation of Fingerprint Images Using Linear Classifier,”
EURASIP J. Applied Signal Processing, vol. 4, pp. 480-494, 2004.
[13] F. Turroni, D. Maltoni, R. Cappelli, and D. Maio, “Improving Fingerprint Orientation Extraction,” IEEE Trans.
Information Forensics and Security, vol. 6, no. 3, pp. 1002-1013, Sept. 2011.
[14] S. Yoon, J. Feng, and A.K. Jain, “On Latent Fingerprint Enhancement,” Proc. SPIE Biometric Technology for
Human Identification VII, pp. 766 707-766 707-10., 2010.
[15] S. Chikkerur, A.N. Cartwright, and V. Govindaraju, “Fingerprint Enhancement Using STFT Analysis,” Pattern
Recognition, vol. 40, no. 1, pp. 198-211, 2007.

Enhanced Latent Fingerprint Segmentation through Dictionary Based Approach

Enhanced Latent Fingerprint Segmentation through Dictionary Based Approach

More Related Content

What's hot (17)

Similar to Enhanced Latent Fingerprint Segmentation through Dictionary Based Approach (20)

More from Editor IJMTER (20)

Recently uploaded (20)

Enhanced Latent Fingerprint Segmentation through Dictionary Based Approach