Adjacent Orientation Vector Based Fingerprint Minutiae Matching System1 G. S. Ng1*, X. Tong2, X. Tang2, D. Shi1 School of Computer Engineering, Nanyang Technological University, Singapore 639798 2 Department of Computer Science and Engineering, Harbin Institute of Technology, Harbin, China * Corresponding author, Email:
[email protected] 1
Abstract Minutia matching is the most popular approach to fingerprint recognition. In this paper, we analyzed a novel fingerprint feature named adjacent orientation vector, or AOV, for fingerprint matching. In the first stage, AOV is used to find possible minutiae pairs. Then one minutiae set is rotated and translated. This is followed by a preliminary matching to ensure reliability as well as a fine matching to overcome possible distortion. Such method has been deployed to a payroll and security access information system and its workability is encouraging. The information system aims to offer a highly secured and automated identification system for payroll tracking as well as authorized access to working areas. 1. Introduction Because of uniqueness, as a personal identification method, fingerprint has been widely used in the past decades. The most popular matching strategy for fingerprint identification is minutiae matching [1-3]. The simplest pattern of the minutiae-based representations consists of a set of minutiae, including ridge endings and bifurcations defined by their spatial coordinates. Each minutia is described by its spatial location associated with the orientation. Although a set of minutiae has been widely used for matching, the noise problem in a fingerprint image has not been solved. The disadvantage of minutiae based method is the lack of robustness, there are some alternative methods proposed, for instance, Jain’s filterbank method [4] and Isenor and Zaky’s graph matching method [5]. The feature vector of minutia generally consists of the minutia type, the coordinates and the tangential angle of the minutia. The automatic fingerprint verification/identification is then achieved with a kind of point pattern matching instead of the fingerprint image matching. Several point pattern matching algorithms have been proposed in the literature [6-7]. Some comments on these algorithms are given in [8]. The point pattern matching is generally intractable because the correspondences between the two point sets of template and input fingerprint are unknown. The minutia correspondences are difficult to obtain due to several 1
factors such as the rotation, translation and deformation of the fingerprints, the location and direction errors of the detected minutiae as well as the presence of spurious minutiae and the absence of genuine minutiae. Many researchers have tried to make the minutiaebased method robust. Ross [9] describes an algorithm using both minutiae and texture features. Ratha and Karu [1] use Hough transform based matching method. By specifying the scale, rotation and shift parameters, a Hough transform is conducted on a minutiae set. A score can be obtained by using these three parameters. The maximal score is used for the best fingerprint. Bhanu and Tan [2] presents a model-based indexing scheme of continuous classification based on novel features of triangles formed by the triplets of minutiae as the basic representation unit. The triangle features used are angles, handedness, type, direction, and maximum side. The experimental results show that this approach can greatly reduce the retrieved fingerprints for further verification in the presence of translation, rotation, scale, shear, occlusion, and clutter. Jiang and Yau [3] proposed a minutia matching technique using both local structure and global structure of minutiae. While comparing two fingerprints, a human expert used to manually examine the local positional relations between minutiae (i.e. the local structure of the minutiae) and to confirm the verification using the fingerprint unique global position structure that consists of the whole minutiae set. The local structure of a minutia is rotation and translation invariant because it consists of the direction and location relative to some other minutiae. Moreover, the local structure can tolerate reasonable deformation because it is formed from only a small area of the fingerprint. So the local structures can be directly used for matching and the best matched local structures provide the correspondences for aligning the global structure of the minutiae. The aligned global structure together with the result of the local structure matching finally determines whether the two fingerprints are generated from the same finger. As we know, adjacent feature of a minutia is very important for matching. Jiang and Yau [3] take advantage of adjacent minutiae but there are two drawbacks of their method. In the first place, it is time consuming to find adjacent minutiae and to build a local data structure. Secondly, there will be various adjacent relations between
This research is supported by the National Natural Science Foundation of China under Grant No. 60332010.
0-7695-2128-2/04 $20.00 (C) 2004 IEEE
minutiae that lead to the complexity of data structure. In this paper, we systematically analyze the work mechanism of a novel fingerprint feature called Adjacent Orientation Vector (AOV)[10]. The AOV is easy to be extracted and the matching between AOVs is not that complex. More importantly, it has a fixed data structure. A fine matching strategy makes the algorithm more tolerant to fingerprint distortion. The paper is organized as follows: Section 2 describes the AOV method in detail. An information system for payroll and security access that uses this method is presented Section 3. Section 4 concludes our work. 2. Adjacent orientation vector Figure 1 illustrates Adjacent Orientation Vector (AOV). If a is a minutia of a fingerprint, then b is the corresponding orientation point, whereas c, d, e and f are the four adjacent minutiae satisfying ac = ad = ae = af = Ddis , ∠bac = 0 , ∠bad = π ,
2.1. Minutiae pair searching
Let < o11 , o12 , o13 , o14 , n11 , n12 , n13 , n14 > and < o21 , o22 , o23 , o24 , n21 , n22 , n23 , n24 > be two AOVs. In order to compare the two AOVs, we need to know the similarity between them. The following formula gives a matching score of two AOVs. AOV_Score = fo (|o11-o12|o+|o21-o22|) + fo (|o13-o14|+|o23-o24| ) (1) + fn (|n11-n12|+|n21-n22| ) + fn (|n13-n14|+|n23-n24| ) where fo is a factor for orientation and fn is a factor for ridge number. Assume that there are two sets of minutiae points P and Q extracted from two fingerprint images using Equation 1, all possible matched pairs < pm , qn > can be found, where pm ∈ P and qn ∈ Q .
∠bae = π , and ∠baf = 3π / 2 where Ddis is a constant. Since a, b, c and d are four points in the fingerprint image, there will be certain orientation at the four points. We assume the difference between the four orientations and minutiae orientation are o1 , o2 , o3 and o4 respectively. The ridge number between ac, ad, ae and af is also used to ensure reliable matching, which is named as n1 , n2 , n3 and n4 respectively. The vector < o1 , o2 ,
2.2. Matching score calculation
o3 , o4 , n1 , n2 , n3 , n4 > is named AOV. As we know, if p and q are corresponding minutiae from two images of the same finger, p and q will have very similar adjacent features. Therefore, p and q will have very similar AOV. Compared with Jiang and Yau’s local features [3], the AOV is easier to be extracted. Because of its simple structure, the matching between two fingerprint images is faster.
matching and Tp is a threshold for preliminary matching
In this section, two matching algorithms will be described, i.e. a preliminary matching to ensure reliability and a fine matching to overcome possible distortion. 2.2.1. Preliminary matching
In Figure 2, S g is matched pair set in preliminary
Step 1: S g =NULL Step 2: PreScore=0 for all p ∈ P and q ∈ Q { if < p, q >∈ S m { PreScore=PreScore+1 } } if (PreScore > T p ){ add < p, q > to sets S g }
score. Figure 2. Preliminary matching algorithm. 2.2.2. Fine matching
Figure 1. Illustration of adjacent orientation vector.
However, the above prelimnary algorithm can not handle distortion. Hence we need a fine matching that makes the algorithm more robust. The fine matching algorithm is shown in Figure 3. In that algorithm, we use two thresholds T y and T θ to determine whether a minutiae pair can be added to S g or not.
0-7695-2128-2/04 $20.00 (C) 2004 IEEE
3. Payroll and security access system based on fingerprint matching
Step 1: Given < i, j >∈ S g and < s, t >∉ S g
(
) (
if ( i y − s y − j y − t y
A payroll and security access system has been implemented using the fingerprint minutia matching technique. Such system is to provide efficient networked services to an intranet for accessing the working area as well as providing payroll tracking. The information system consists of 3 main components: Register Module, Attendance Log In Module and a web-based Payroll Resource Control Module. A fingerprint image reader is used to acquire fingerprint images and a control interface is developed to incorporate the reader into the information system. Figure 4 shows an overview of such system using client-server model.
) < T and
( iθ − sθ ) − ( jθ − tθ )
y
< Tθ ){
add < s, t > to S g } Step 2: Repeat Step 1 until no minutiae pair can be added to S g .
Figure 3. Fine matching algorithm.
Control commands
Control commands
Fingerprint image
Client
Fingerprint image
Fingerprint Reader Control Interface
Fingerprint matching result
Fingerprint image
Server Figure 4. The Client-server model for payroll and security access. system.
3.1. Register Module
The Register Module is a user application that handles registration of staffs’ information into the database and also the collection of fingerprint sample images from the fingerprint reader. This module resides on the client machine and makes use of the shared services provided by the server for making database queries.
systems such as using punch cards or smart cards, this module offers a more efficient and economical solution. During the polling, the fingerprint reader may capture a blank screen. Thus, the module must be able to differentiate between an image that is a blank screen and one with a valid fingerprint (see Figure 5). Otherwise, extra processing cost will be incurred for a blank image submitted to the fingerprint matching at the server.
3.2. Attendance Log In Module
The Attendance Log In Module is responsible for authorizing access and recording of staff attendance by using fingerprint minutia matching technique. This module is also linked to the Payroll Resource Control Module because it handles staff attendance and keeps track of the number of man-hours clocked daily by each individual staff. As the staff enters or leaves working area, this module logs the time and it is used to calculate the working hours of the staff. Compared to existing
(a) (b) Figure 5. (a) A blank screen with some dirts. (b) A valid fingerprint image.
0-7695-2128-2/04 $20.00 (C) 2004 IEEE
3.3. Payroll Resource Control Module
The web-based Payroll Resource Control Module consists of three categories for 3 types of users: payroll administrator, human resource administrator and the security administrator. The payroll administrator is able to access any information pertaining to the payroll whilst the human resource administrator has access to employees’ personal information. The security administrator will be responsible for maintenance of the database and the registration of new users including their fingerprints. Figure 6 shows the functionalities of this module.
[4]
[5]
[6]
4. Conclusions
In this paper, a novel fingerprint feature called adjacent orientation vector (AOV) is systematically analyzed. By predicting possible minutiae pair, AOV helps in the matching procedure. A preliminary matching is employed to find all the possible matched pairs. Considering possible distortion, a fine matching is then used to make fingerprint matching more robust. This has been verified by applying the above method on a payroll and security access system. References
[1]
[2]
[3]
[7]
[8]
[9]
N. K. Ratha and K. Karu, “A real time matching system for large fingerprint databases,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 18, no. 8, pp. 799-813, 1996. B. Bhanu and X. Tan, “Fingerprint indexing based on novel features of minutiae triplets,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 25, no. 5, pp. 616-622, 2003. X. Jiang and W. Y. Yau, “Fingerprint
[10]
minutiae matching based on the local and global structures,” Proc. ICPR2000, vol. 2, pp. 1042-1045, Sept. 2000. A. K. Jain, S. Prabhakar, L. Hong, and S. Pankanti, “Filterbank-based fingerprint matching,” IEEE Transactions on Image Processing, vol. 9, no. 5, pp. 846-859, 2000. D. K. Isenor and S. G. Zaky, “Fingerprint identification using graph matching”, Pattern Recognition, vo. 19, no. 2, pp. 113-122, 1986. A. Ranade and A. Rosenfeld, “Point pattern matching by relaxation,” Pattern Recognition, vol. 12, no. 2, pp. 269-275, 1993. J. Starink and E. Backer, “Finding point correspondence using simulated annealing,” Pattern Recognition, vol. 28, no. 2, pp. 231-240, 1995. A. K. Jain, L. Hong and R. Bolle, “OnLine Fingerprint Verification,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 4, pp. 302-314, 1997. A. Ross, “Fingerprint matching using minutiae and texture features,“ Proceedings of International Conference on Image Processing, pp. 282-285, 2001. X. Tong and X. Tang, “Fingerprint minutiae matching based on adjacent orientation,” Proceedings of the 7th International Conference for Young Computer Scientists, pp. 465-469, August 8-10, 2003.
Web-based Payroll Resource Control Module
Human Resource
Payroll
Security
Attendance report
Update record
Add user
Payroll report
Delete record
Update user
View record
Delete user View user
Figure 6. Functionalities of web-based Payroll Resource Control module.
0-7695-2128-2/04 $20.00 (C) 2004 IEEE
