ABSTRACT

Overlapped Fingerprint Recognition is a complex pattern recognition problem. It is

difficult to design accurate algorithms capable of separating the overlapped fingerprint,

extracting salient features and matching them in a robust way. Latent fingerprints lifted from

crime scenes often contain overlapping prints, which are difficult to separate and match by

state-of-the-art fingerprint matchers. A few methods have been proposed to separate

overlapping fingerprints to enable fingerprint matchers to successfully match the component

fingerprints. These methods are limited by the accuracy of the estimated orientation field,

which is not reliable for poor quality overlapping latent fingerprints. In this proposed

method, we try to improve the robustness of overlapping fingerprints separation, particularly

for low quality images. Our algorithm reconstructs the orientation fields of component prints

by modelling fingerprint orientation fields and then correcting it using dictionary based

approach. In order to facilitate this, we utilize the orientation cues of component fingerprints,

which are manually marked by fingerprint examiners. This additional mark-up is acceptable

in forensics, where the first priority is to improve the latent matching accuracy. The proposed

orientation field estimation algorithm consists of an offline dictionary construction stage and

an online orientation field estimation stage.

In the offline stage, a set of good quality fingerprints of various pattern types (arch, loop,

and whorl) is manually selected and their orientation fields are used to construct a dictionary

of orientation patches. In the online stage, given a fingerprint image, its orientation field is

automatically estimated using model based and dictionary based approach.

The proposed method will not only work on simulated overlapping prints, but also on real

overlapped latent fingerprint images. The proposed algorithm can be more effective in

separating poor quality overlapping fingerprints and enhancing the matching accuracy of

overlapping fingerprints.

1

1. INTRODUCTION

Fingerprint recognition or fingerprint authentication refers to the automated method of

verifying a match between two human fingerprints. Fingerprints are one of many forms of

biometrics used to identify an individual and verify their identity. Because of their

uniqueness and consistency over time, fingerprints have been used for over a century, more

recently becoming automated (i.e. a biometric) due to advancement in computing

capabilities.

Typically, the input image contains only a single fingerprint. However, in practice,

particularly in forensics, two or more fingerprints could overlay on top of each other,

resulting in an overlapped fingerprint image. Available fingerprint matchers, however,

cannot accurately match overlapping fingerprints, because they assume that a fingerprint

image contains only a single fingerprint and hence single orientation field. Our interest here

is to develop algorithms to separate overlapping latents that will serve as a valuable tool in

forensics. Note that in forensics, the matching accuracy of latents is extremely critical even if

it involves some degree of manual intervention by latent examiners, including manual

markup.

1.1 Objective of the project

The objective of this research is to find a robust solution for separating the overlapped

fingerprint particularly for latent fingerprint which is difficult to separate using existing

techniques.

1.2 Scope of the project

The proposed method will serve a valuable tool in forensics for separation of

overlapped simulated fingerprint, inked fingerprint and also on latent overlapped fingerprints.

2. LITERATURE SURVEY

Fan et al. [3] proposed an algorithm to separate overlapped fingerprints based on image

enhancement using manually marked orientation field. However, it is very tedious and time-

consuming for the user to manually mark the orientation field of each component fingerprint

in the overlapped fingerprint image. Geng et al. [4] proposed to use morphological

2

component analysis to separate overlapped fingerprints. However, their experiment shows

that this algorithm can only separate the component fingerprint which dominates the

overlapped image. chen et al. [1] In this paper present an algorithm to separate overlapped

fingerprints: The algorithm consists of three steps: 1) Initial orientation field is estimated

using local Fourier analysis. 2) Relaxation labeling method is employed to label the initial

orientation field into two classes. 3) Two component fingerprints are separated by enhancing

the overlapped fingerprint image using Gabor filters tuned to these two component

orientation fields.

Problem with approach is it does not perform very well when the singularity region of

component fingerprint overlapped because the relaxation algorithm solely based on local

continuity of orientation field.

shi et al. [2] propose overlap orientation field method based on constrained relaxation

labeling similar to chen et al. [1]. This algorithm is differing from chen et al[1] in following

aspects.1) Utilization of non-overlapped area. In this work non-overlapped area is utilized as

important constraints during relaxation labeling process. 2) Mutual exclusion constraint. We

treat each overlapped block as a single object rather than two ones to strictly enforce the

mutual exclusion constraint, namely, two candidate orientations in an overlapped block

cannot belong to the same fingerprint. 3) Order of updating label probabilities. We

sequentially update label probabilities in an overlapped block in an ascending order of the

distance between itself and non overlapped area. While in [1], the label probabilities in all

overlapped blocks are updated in parallel. Feng et al. [10] As this paper is similar to shi et al.

[2] but the difference is improved versions of the basic algorithm for two special but frequent

cases: 1) The mated template fingerprint of one component fingerprint is known and

2) The two component fingerprints are from the same finger.

2.1.8 zhao et al.[3] This approach is almost overcome the drawbacks of relaxation labelling

technique. In this paper, instead of separating the estimated mixed orientation field, they

reconstruct the orientation fields of component fingerprints via modelling orientation fields

and then predicting unknown orientation fields based on a small number of manually marked

orientation cues in fingerprints. This model based method significantly improves the

3

accuracy of overlapping fingerprints separation, especially for the practical scenario of poor

quality overlapped latent images.

3. PROPOSED METHODOLOGY

The proposed orientation field estimation algorithm consists of an offline dictionary

construction stage and an online orientation field estimation stage illustrated in fig. 3.1.

In the offline stage, a set of good quality fingerprints of various pattern types (arch, loop,

and whorl) is manually selected and their orientation fields are used to construct a dictionary

of orientation patches. In the online stage, given a fingerprint image, its orientation field is

automatically estimated. The figure 3.1 shows the online and offline stage in detail.

Figure.3.1 Proposed system Overview

3.1 Offline Dictionary construction

The dictionary consists of a number of orientation patches of the same size. The numbers of

orientation patches, whose orientation elements are all available, are obtained by sliding a

window.

The greedy algorithm for dictionary construction is described below in Table 3.1:

4

Reference Fingerprints

Reference Orientation

fields

Input Fingerprint

Initial Orientation

Field

Dictionary of orientation

patches

Corrected orientation field

Orientation

Field Estimation

Dictionary

Construction

Orientation Estimation

DictionaryLookup

Context based correction

Online

Offline

Gabor

Filter

Step Description1 The first orientation patch is added into the dictionary, which is initially empty.

2 Then we test whether the next orientation patch is sufficiently different from all the orientation patches in the dictionary. If yes, it is also added into the dictionary; otherwise, the next orientation patch is tested

3 Repeat step 2 until all orientation patches has been tested.Table 3.1 Dictionary Construction.[11]

3.2 Online stage

In the online stage, given a fingerprint image, its orientation field is automatically estimated,

the steps for online is described in table 3.2.

Step Description1 Initial estimation. The initial orientation field is obtained using a Model based

technique [9].2 Dictionary lookup. The initial orientation field is divided into overlapping

patches. For each initial orientation patch, its six nearest neighbours in the dictionary are viewed as candidates for replacing the noisy initial orientation patch.

3 Context-based correction. The optimal combination of candidate orientation patches is found by considering the compatibility between neighbouring orientation patches.

Table 3.2. Online Stage.

3.2.1 Initial orientation field estimation of two component fingerprint

The initial orientation field is obtained using a Model based technique [9]. The following

parameters are used in algorithm to find the orientation field.

DescriptionRegion Segmentation by manual marking singular points, orientation cues and overlapped region.

Modelling orientation field using combination model of zero pole and polynomial model. Zero pole is use to detect area near singular point and polynomial model is used to detect remaining area. The zero pole model is

5

We first remove the influence of singular points from the orientation field by subtracting from, and then approximate the residual orientation field with a set of basis functions.

(2)

The cosine and sine components of the doubled residual orientations can be approximated by

(3)

Given the residual orientation field in the region of interest, the coefficients can be obtained by solving the following minimization problems using least squares optimization.

(4)

Step 3) The residual orientation field can be calculated as:

(5)

Finally, the orientation at is obtained by adding back the influence of singular points to the estimated residual orientation, i.e.,

(6)

3.2.2 Reconstruct the orientation field using algorithm

In the reconstructing process the orientation field is reconstructed using combination model.

The zero pole and polynomial model is used as a combination model.

Steps DescriptionInput: Region of interest; θ(ΩC): Orientation cues in ΩC ;SP: Singular points

6

Output:

: Reconstructed Orientation field in Ω

1 Compute the orientation fields of the singular point according to(1):

2 Compute the residual orientation fields in ΩC:

3 Initialize the prediction area:

4 While is not empty do

5 Estimate the model coefficients based on according to (4)

6 Compute the predicted residual orientations in according to (5):

7 Regularize the residual orientations in :

8 ,

9 end while

10 Compute the reconstructed orientation field according to (6):

.

Table 3.3 Reconstructions process [9]

3.3 Dictionary lookup

The similarity between an initial orientation patch and a reference orientation patch is

computed by comparing corresponding orientation elements. Hence, a candidate is selected

from initial candidates using the following greedy strategy described in table 3.4.

Step Description

1 Choose the first initial candidate.

2 The next initial candidate is compared to each of the chosen candidates. If its

similarity to all the chosen candidates is below a predefined threshold it is chosen.

7

It is defined as: S(Θ,ɸ)=ns/nf.

3 Repeat step 2 for all the initial candidates until candidates have been chosen or all

initial candidates have been checked

Table 3.4 Dictionary Lookup

3.4 Context-based orientation field correction

After dictionary lookup, we obtain a list of candidate orientation patches, for an initial

orientation patch. To resolve the ambiguity, i.e., determine a single candidate for each patch,

contextual information needs to be utilized. We address this problem by searching for a set of

candidates, r, which minimizes an energy function. We consider two factors in designing the

energy function: 1) The similarity between the reference orientation patches and the

corresponding initial orientation patches, and 2) The compatibility between neighboring

reference orientation patches. The energy function is defined as:

Where Es denotes the similarity term, Ec denotes the compatibility term, and Wc is the weight of compatibility term shown in table 3.5.

The similarity term is defined as:

The compatibility term is defined as:

The compatibility is computed as:

Table 3.5 Context based correction [11]

After this step of context based correction we get the orientation field of two component

fingerprint.

3.5 Separating overlapped fingerprint

8

Two important parameters of 2D Gabor filters are local ridge orientation and frequency.

Based on the component ridge orientation field, we estimate the ridge frequency map of each

component fingerprint using the method proposed in [5]. When the ridge orientation field and

ridge frequency map are obtained, Gabor filtering can connect broken ridges and remove

intervening ridges. The main steps of the algorithm include: 1) Local orientation estimation:

The orientation image is estimated from the normalized input fingerprint image. 2) Local

frequency estimation: The frequency image is computed from the estimated orientation

image.3) Filtering: A bank of Gabor filters which is tuned to local ridge orientation and ridge

frequency is applied to the ridge-and-valley pixels in the normalized input fingerprint image

to obtain an enhanced fingerprint image. This is the proposal which can be efficiently used

for reconstruction and prediction model more accurately and precisely.

4. CONCLUSION

The robust and efficient overlapped fingerprint separation has been proposed .As finger print

orientation is very important step in separation algorithm; we have proposed a robust

orientation field estimation algorithm for latent fingerprint enhancement. There are several

advantages of the proposed fingerprint separation method. The method proposed not only

effective for simulated and inked fingerprint but also effective for latent fingerprint

separation. In the proposed system we are using model based technique and dictionary based

approach for initial orientation estimation which is inspired from spelling correction

techniques in natural language processing, as state of art method of relaxation labeling

algorithm used local fourier analysis which was the bottle neck. So here we are using model

based orientation estimation technique. The proposed algorithm reconstructs the orientation

field of overlapping fingerprints based on a set of manually marked features, including

regions of interest, singular points, and orientation cues. Based on the underlying model of

fingerprint ridge orientation field, the proposed method can simultaneously predict unknown

orientations in fingerprints and use the dictionary based approach for context based

correction. and finally the two component fingerprints are separated by filtering the

overlapped fingerprint image using Gabor filters tuned to the component orientation fields.

9

REFERENCES

[1] F. Chen, J. Feng, A. K. Jain, J. Zhou, and J. Zhang, “Separating overlapped fingerprints,”

IEEE Trans. Inf. Foren. Secur., vol. 6, no. 2, pp. 346–359, 2011.

[2] Y. Shi, J. Feng, and J. Zhou, “Separating overlapped fingerprints using constrained

relaxation labeling,” in Proc. Int. Joint Conf. Biometrics, 2011.

[3] X. Fan, D. Liang, and L. Zhao, “A scheme for separating overlapped fingerprints based

on partition mask,” (in Chinese) Comput. Eng. Applicat., vol. 40, no. 2, pp. 80–81, 2004.

[4] R. Geng, Q. Lian, and M. Sun, “Fingerprint separation based on morphological

component analysis,” (in Chinese) Comput. Eng. Applicat., vol. 44, no. 16, pp. 188–190,

2008.

[5] L. Hong, Y. Wan, and A. K. Jain, “Fingerprint image enhancement: Algorithm and

performance evaluation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 20, no. 8, pp. 777–

789, 1998.

[6] J. Zhou and J. Gu, “A model-based method for the computation of fingerprints

orientation field,” IEEE Trans. Image Process., vol. 13, no. 6, pp. 821–835, 2004.

[7] J. Gu, J. Zhou, and D. Zhang, “A combination model for orientation field of fingerprints,”

Pattern Recognit., vol. 37, pp. 543–553, 2004.

[8] J. Feng and A. K. Jain, “Fingerprint reconstruction: From minutiae to phase,” IEEE

Trans. Pattern Anal. Mach. Intell., vol. 33, no. 2, pp. 209–223, 2011.

[9] Q. Zhao and A.K. Jain, “Model Based Separation of Overlapping Latent Fingerprints,”

IEEE Trans. Information Forensics and Security,vol. 7, no. 3, pp. 904-918, June 2012.

10

[10] J. Feng, Y. Shi, and J. Zhou, “Robust and Efficient Algorithms for Separating Latent

Overlapped Fingerprints,” IEEE Trans. Information Forensics and Security, 2012.

[11]Jianjiang Feng, Member, IEEE, Jie Zhou, Senior Member, IEEE, and Anil K. Jain,

“Orientation Field Estimation for Latent Fingerprint Enhancement” IEEE transactions on

pattern analysis and machine intelligence, vol. 35, no. 4, pp. 925-940, April 2013.

11