Handwritten Signature Verification by Multiple Reference Sets

Size: px

Start display at page:

Download "Handwritten Signature Verification by Multiple Reference Sets"

Laureen Ross
6 years ago
Views:

1 Handwritten Signature Verification by Multiple s D. Impedovo (^)( ), R. Modugno (*)( ),G. Pirlo (*)( ) (*) ( ), E. Stasolla (*) Dipartimento di Informatica Università degli Studi di Bari Via Orabona 4 Bari Italy (^) Dipartimento di Elettrotecnica ed Elettronica Politecnico di Bari Via Orabona 4 Bari Italy ( ) Centro Rete Puglia - Università degli Studi di Bari Via G. Petroni 15/F.1 Bari Italy {segreteria@retepuglia.uniba.it} Abstract This paper presents a new approach for on-line handwritten signature verification, which exploits the potential of multiple reference sets. Preliminarily, system performance is estimated using different sets of reference signatures for each writer. Successively, reference sets leading to diverse system behaviors are enrolled into the personal nowledge-base and used in a multi-stage verification process. The experimental results show the effectiveness of the proposed approach, compared to traditional techniques. Keywords: Biometry, Classification, Dynamic Signature, Signature Verification, Multi-expert system. 1. Introduction The development of the e-society is strongly dependent on the possibility to fully exploit potentials of internetworing. In other words it is necessary for more and more people to use the net for performing all the daily activities and operations, such as teamwor cooperation and distance learning, baning transactions and fund transfers, access to information resources, etc. Therefore there is a growing need for secure personal verification. Automatic personal verification systems can be categorized on the basis of the means they use for personal verification: physical mechanisms belonging to the individual (i.e. ey or badge), information (i.e. password, numeric string, ey-phrase) and biometric characteristics (i.e. speech, finger-print, palm-print, signature) [1, 2, 3]. Compared to personal verification systems based on physical mechanisms and information, the main advantage of biometric systems is that biometric characteristics cannot be lost, stolen or forgotten [4, 5]. Furthermore, among the various biometrics, handwritten signature verification has the advantage that signature has long been established as one of the widespread means for personal verification in our daily life; it is well-recognized by legal and financial institutions and well-accepted by users [1,4,6,7]. Therefore, the field of signature verification has attracted many researchers which are interested not only to the scientific challenges but also to the valuable applications this field offers [8, 9]. Several excellent survey papers report the progress in the field of automatic signature verification [10, 11, 12, 13, 14, 15]. Recently, the extraordinary growth of the internet has augmented the interest toward automatic signature verification, as demonstrated by the creation of specific laws and regulations, approved in many countries and the attention to the standardization of signature data interchange formats by various important associations and institutes [16, 17, 18]. Therefore, along with these important results, signature verification technologies are becoming much more profitable solutions for a wide range of commercial applications such as baning, insurance, health care, ID-security, document management, e-commerce and retail point-of-sale (POS) [19, 20, 21, 22, 23]. Howeve many efforts are still necessary to perform automatic signature verification effectively. In fact, automatic signature verification is a complex tas that involves many biophysical and psychological aspects related to human behavior as well as many engineering issues [24, 25, 26, 27, 28, 29]. A handwritten signature is the result of a complex process based on a sequence of actions stored into the brain and realized by the writing system of the signer (arms and hands) through ballistic-lie movements. Thus, signatures of the same person can considerably differ depending on the physical and psychological condition of the writer. Two types of variability must be considered [9, 10]: short period and long-period variability. Short-period variability is evident on a day-to-day basis, it is mainly due to the psychological condition of the writer and on the writing conditions (posture of the write type of pen and pape size of the writing area, etc.). Longperiod variability is due to the modifications of the physical writing system of the signer as well as of the sequence of actions stored in his/her brain [10, 11]. So fa many efforts have been devoted to automatic signature verification systems using different function-based and parameter-based features as well as advanced techniques for signature comparison [9, 10, 13, 14, 15]. Anyway, whatever feature set and comparison technique is considered, the performance of signature verification systems is strictly dependent on the type and quality of information

2 used for reference and on the way in which it is organized and exploited for verification. So fa two different strategies have been addressed for reference information construction and management [30]. A first strategy uses a single prototype of genuine signatures for each write and several techniques have been proposed for the development of the optimal average prototype for a signer and for determining the optimal threshold representing the personal variability in signing, by shape and dynamic feature combination [30], time- and position-based averaging [31]. A second strategy uses a set of genuine signatures for reference. In this case a crucial problem concerns the selection, among the samples available, of the optimal subset of reference signatures that represents the signature of an individual in the best way. In fact, it is worth noting that the number of authentic signatures, acquired during the enrolment phase, is generally larger than the number of reference signatures employed in the verification process. When static signature verification is considered, the validity of the reference model is evaluated according to specific quality criterion, as for instance the intra-class variability that should be as low as possible [32, 33]. In dynamic signature verification, the selection of the best sub-set of reference signature has been determined on the basis of the stability regions in the signatures of the sub-set, determined by a well-defined index of local stability [34, 35, 36]. Of course, the selection of the best subset of reference signature can be avoided at the cost of using multiple models for signature verification [37, 38, 39], also by considering synthetic signatures generated from the existing ones by convolutions [40], elastic matching [41] and perturbations [42]. This paper presents a new approach for on-line handwritten signature verification, which exploits potential of multiple reference sets leading to diverse system behaviors. In a preliminary phase, system performance is estimated using different subsets of reference signatures extracted from the available set of authentic signatures. Subsets leading to diverse system behavior estimated by means of well-suited optimality functions - are considered for reference and used in a multi-stage verification process. The experimental result shows the potential of the proposed approach with respect to traditional strategies for reference information selection. 2. Analysis of s In this paper the analysis of different sets of genuine signatures is performed on the basis of their effectiveness, when used as reference. The process of signature verification can be considered as a function D( ), that associates a Boolean value to each input signature S i : D(S i )= G if the signature is considered to be genuine; D(S i )= F if the signature is considered to be a Of course, the signature verification process can produce two types of errors: the Type I errors - that concern with the false rejection of genuine signatures, and the Type II errors - that concern with the false acceptance of forged signatures. Therefore, the False Rejection Rate () and the False Acceptance Rate () are the most diffuse estimators of the accuracy of a signature verification system [9, 10]. In many practical applications a trade-off between the two error types must be defined since any reduction of increases, and vice-versa. For this purpose the Total (TER), that is defined as +, is also widely considered as a measure of the overall error of the system [42]. Now, let S r : { S r i i=1,2,,n r } be the set of genuine signatures available for reference and let us consider the problem concerning with the selection of the best subset of reference signatures S, from S r. Traditionally, this can be done by the analysis of the subset S TER for which the TER is minimum (see Figure 1) [42]. In other word let S g = { S g i i=1,2,,n g } and S f = { S f i i=1,2,,n f } be respectively a set of genuine and forgery signatures, the set S TER is considered as the set for which it is minimum the function where: O TER (S ) = ( + ) (1) = 1 N = 1 N g f card card g g D( Si ) = F} f f { S D( S ) = G} i i, (2). (3) Obviously, when S TER is considered, the function D TER( ), that associates to each input signature S i a Boolean value, can be rewritten as: D TER(S i )= G if the signature is considered to be genuine; D TER(S i )= F if the signature is considered to be a In this pape starting from the consideration that the characteristics of different subsets can be exploited to improve the overall verification accuracy, a two-level verification strategy is considered (see Figure 2). At the first level the following subsets are considered: S : the subset for which it is minimum the value of the optimality function O (S ) = α + β, (4) where α and β weight the cost of False Rejection Errors and False Acceptance Errors (in this case α >> β ). In this case the function D ( ), that associates to each input signature S i a Boolean value can be rewritten as: D (S i )= G if the signature is considered to be genuine; D (S i )= F if the signature is considered to be a

3 S : the subset for which it is minimum the value of the optimality function O (S ) = α + β, (5) where α and β weight the cost of False Rejection Errors and False Acceptance Errors (in this case α << β ). In this case the function D ( ), that associates to each input signature S i a Boolean value can be rewritten as: D (S i )= G if the signature is considered to be genuine; D (S i )= F if the signature is considered to be a FFR Figure 1. Single S S TER First Level Second Level S TER* Figure 2. Multiple s S Using S FFR and S the first level allows the direct verification of an input signature S i for the following cases: D (S i )= G and D (S i )= G, then S i is considered to be a genuine sample; D (S i )= F and D (S i )= F, then S i is considered to be a forged sample; Of course, the other cases must be considered through a second-level verification. For this purpose let be S g* = { S g i S g i=1,2,,n g : D (S i ) D (S i )} and S f* = { S f i S f i=1,2,,n f : D (S i ) D (S i )}, the set S TER* is considered, that is defined as the set for which it is minimum the value of the optimality function O TER* (S ) = ( * + * ) (6) where * * 1 * g g g S D( Si ) = " F" } = card (7) g* Card( S ) 1 * f f f S D( Si ) = " G" } = card (8) f * Card ( S ) Obviously, when S TER* is considered, the function D TER*( ), that associates to each input signature S i a Boolean value can be rewritten as: D TER*(S i )= G if the signature is considered to be genuine; D TER*(S i )= F if the signature is considered to be a 3. Signature Verification System The approach proposed in this paper is general and it does not depend on the specificities of the signature verification system considered. Anyway, in this wor the novel approach for multiple reference sets selection has been applied to a online signature verification system that uses very simple strategies based on velocity signals as function-features and Dynamic Time Warping for signature comparison [9, 10, 11]. Although the complete description of the system is not included in this paper (it can be found in the literature [34, 35]), some major characteristics of the system are described in the following. After signature acquisition, the raw data are preprocessed by removing the undesired initial and final sequences of coordinates. Successively the signature is normalised so that the rectangle surrounding the signature is scaled to a fixed area [34]. After preprocessing, the velocity of the tip of the pen during signing is evaluated numerically from the position signal provided by the tablet. Of course, during the training process, the features extracted from the set of reference signatures are enrolled into the personal database. In the comparison phase they are matched against those belonging to the input (test) signature by DTW. The result is used to judge the authenticity of the input signature. For this purpose, the worst dissimilarity measures obtained from the local analysis of the reference signature are used as threshold

4 values, since they are considered as a measure of personal variability in signing. Depending on the approach used for reference set selection, the following verification schemes are adopted: 1) Single reference set. In this case the verification decision for an input signature S i is performed by the traditional one-level scheme: D TER(S i )= G S i genuine D TER(S i )= F S i 2) Multiple reference sets. In this case the verification decision for an input signature S i is performed by the new two-level scheme: (first level) D FFR(S i )= G and D (S i )= G S i genuine D FFR(S i )= F and D (S i )= F S i otherwise (second level) D TER*(S i )= G S i genuine D TER*(S i )= F S i 4. Experimental Results For the experimental tests, a database has been developed by ten signers, that have collected the genuine signatures, and ten writers, that have produced the forged samples in daily writing sessions. For each write the database includes forty genuine signatures and forty silled forgeries. Writers have had time to exercise and improve their capabilities in imitating the genuine signatures. Furthermore, each signer has produced N r =10 extra signatures to be used for the selection of the best subsets of reference signatures. Of course, although the proposed approach is general, at the moment only subsets of =3 signatures have been considered. Therefore, in the analysis of the reference sets N r 10 = =120 sets of =3 3 signature each have been considered. Table I compares the performance of the system when the single reference set selected by (1) is considered (Approach 1) and when multiple reference sets selected by (4) are used (Approach 2). In the last case the woring parameters are the following: α =1, β =10, (eq. 5); α =10,β =1 (eq. 6). Although a small testing database has been considered so fa the preliminary result point out that, the approach based on multiple reference sets generally outperforms the traditional approach based on single reference. Concerning and, the new approach is superior to the traditional one respectively 9 times out of 10 and 7 times out of ten. At the best, a 100% reduction has been obtained for the of the signers n. 5, 7, 8, 9, 10; and for the of the signers 8, 9, 10. If the total system error rate - defined as + - is considered, an improvement ranging from 11.7% to 100% has been obtained set 9 times out of Conclusion This paper presents a new approach for selecting multiple sets of reference specimens for automatic signature verification. Form the analysis of different behaviours related to various sets of reference signatures, a multi-level verification strategy is proposed which uses well-selected sets of reference signatures. The experimental results, carried out in the field of dynamic signatures, demonstrate the effectiveness of the new approach. References [1] C. Vielhauer and J. Dittmann, Biometrics for User Authentication : Encyclopedia of Multimedia, ed. B. Furth, Springer-Verlag, Berlin, [2] A. Jain, R. Bolle, S. Pananti (eds), Biometrics: Personal Identification in Networed Society, Kluwer Academic Publishers, Boston, USA, [3] S. Nanavati, M. Thieme, R. Nanavati, Biometrics: Identity Verification in a Networed World, pp , Wiley, New Yor, [4] K.W. Boye V. Govindaraju, N.K. Ratha (Eds.), Special Issue on Recent Advances in Biometric Systems, IEEE Trans. on Syst., Man and Cybernetics Part B, Vol. 37, No. 5, Oct [5] S. Prabhaa J. Kittle D. Maltoni, L. O'Gorman, T. Tan (Eds.), Special Issue on Biometrics: Progress and Directions, IEEE T-PAMI, Vol. 29, No. 4, April [6] M.C. Fairhurst, New Perspectives in Automatic Signature Verification, Information Security Technical Report, Vol. 3, No. 1, 1998, pp [7] M.C. Fairhust, S. Ng, Management of access through biometric control: A case study based on automatic signature verification, Universal Access in the Information Society (UAIS), Springer-Verlag, Vol. 1, No. 1, 2001, pp [8] E. Newham, Survey: Signature Verification Technologies, Signature Verification, Elsevier Science Ltd, April 2000, pp [9] R. Plamondon, G. Lorette, Automatic Signature Verification and Writer Table I. Performance. Signer Approach 1: Single 17% 20% 18% 25% 16% 16% 9% 2% 8% 18% 17% 20% 18% 25% 16% 16% 9% 2% 8% 18% + 34% 40% 36% 50% 32% 32% 18% 4% 16% 36% Approach 2: Multiple s 5% 2.5% 22.5% 7.5% 0% 5% 0% 0% 0% 0% 25% 40% 5% 20% 20% 5% 5% 0% 0% 0% + 30% 42% 27.5% 27.5% 20% 10% 5% 0% 0% 0%

5 Identification - The State of the Art, Pattern Recognition, 1989, Vol 22(2), pp [10] F. Leclerc, R. Plamondon, Signature verification: The state of the Art , IJPRAI, 1994, Vol 8(3),pp [11] G. Dimauro, S. Impedovo, G. Pirlo, Algorithms for Automatic Signature Verification, in Handboo of Character Recognition and Document Image Analysis, World Scientific Publ., H. Bune and P. S. P. Wang (Eds.), 1997, pp [12] G. Pirlo, D. Impedovo, Automatic Signature Verification: The State of the Art, IEEE-System Man and Cybernetics Part C, 2008 (in press). [13] V. S. Nalwa, Automatic On-line Signature Verification, Proceedings of the IEEE, Vol. 85, No. 2, Feb. 1997, pp [14] G. Dimauro, S. Impedovo, M.G. Lucchese, R. Modugno, and G. Pirlo, Recent Advancements in Automatic Signature Verification, 9 th International Worshop on Frontiers in Handwriting Recognition (IWFHR-9), Kichijoji, October 25-29, [15] R. Plamondon, S.N. Srihari, On-line and off-line handwriting recognition: A Comprehensive Survey, IEEE T-PAMI, 2000, 22(1), pp [16] ANSI, Information technology Biometric Data Interchange Formats Signature/Sign Data, ANSI INCITS , [17] ISO, Information technology Biometric data interchange formats Part 7: Signature/sign time series data, ISO/IEC FCD [18] ISO Information technology Biometric data interchange formats Part 11: Signature/Sign Processed Dynamic Data, ISO/IEC WD [19] M.C. Fairhurst, Signature Verification Revisited: Promoting Practical Exploitation of Biometric Technology, IEE Electronics and Communication Engineering Journal (ECEJ), Vol. 9, No. 6, Dec. 1997, pp [20] U. Uludag, S. Pananti,S. Prabhaa A.K. Jain, Biometric Cryptosystems: Issues and Challenges, Proc. of the IEEE, Vol. 92, No. 6, June 2004, pp [21] O. Urèche and R. Plamondon, Document Transport, Transfe and Exchange, Security and Commercial Aspects, Proc. ICDAR 99, Bangalore, India, Sept. 1999, pp [22] C. Vielhaue L.C. Ferri, Applications of a hologram watermaring protocol: aging-aware biometric signature verification and time validity chec with personal documents, Proc. SPIE, Vol. 5020, Jun. 2003, pp [23] C. Vielhaue A Behavioural Biometrics, Public Service Review: European Union, 2005, No. 9, pp [24] J.-J. Brault and R. Plamondon, A Complexity Measure of Handwritten Curves: Modeling of Dynamic Signature Forgery, IEEE Transactions on Systems, Man, and Cybernetics (T-SMC), Vol. 23, No. 2, 1993, pp [25] G. Pirlo, Algorithms for Signature Verification, in Proc. of NATO-ASI Series Fundamentals in Handwriting Recognition, S. Impedovo (Ed.), Springer-Verlag, Berlin, 1994, pp [26] R. Plamondon, A Kinematic Theory of Rapid Human Movements: Part I: Movement Representation and generation, Biological Cybernetics, Vol. 72, No. 4, 1995, pp [27] R. Plamondon, A Kinematic Theory of Rapid Human Movements: Part II: Movement Time and Control, Biological Cybernetics, Vol. 72, No. 4, 1995, pp [28] R. Plamondon, A Kinematic Theory of Rapid Human Movements: Part III: Kinetic Outcomes, Biological Cybernetics, Jan [29] R. Plamondon, W. Guerfali, The generation of handwriting with deltalognormal synergies, Biological Cybernetics, 1998, Vol. 78 (2), pp [30] C. Schmidt and K.-F. Kraiss, Establishment of Personalized Templates for Automatic Signature Verification, 4 th International Conference on Document Analysis and Recognition (ICDAR-4), IEEE Computer Society, Ulm, Germany, Vol. I, Aug. 1997, pp [31] B. Wirtz, Average Prototypes for Stroe-Based Signature Verification, 4 th International Conference on Document Analysis and Recognition (ICDAR-4), IEEE Computer Society Press, Ulm, Germany, Vol. I, Aug. 1997, pp [32] C. Allgrove and M.C. Fairhurst, Enrolment model stability in static signature verification, 7 th International Worshop on Frontiers in Handwriting Recognition (IWFHR-7), L.R.B. Schomaer and L.G. Vuurpijl (Eds.), Amsterdam, The Netherlands, September 2000, pp [33] M.C. Fairhurst, Signature Verification Revisited: Promoting Practical Exploitation of Biometric Technology, IEE Electronics and Communication Engineering Journal (ECEJ), Vol. 9, No. 6, Dec. 1997, pp [34] G. Congedo, G. Dimauro, A.M. Forte, S. Impedovo, G. Pirlo, Selecting Reference Signatures for On-Line Signature Verification, 8 th International Conference on Image Analysis and Processing (ICIAP-8), Series: Lecture Notes in Computer Science, Vol. 974, Springer-Verlag Berlin, Heidelberg, C. Braccini, L. De Floriani and G. Vernazza (Eds.), San Remo, Italy, Sept. 1995, pp [35] V. Di Lecce, G. Dimauro, A. Guerriero, S. Impedovo, G. Pirlo, A. Salzo, L. Sarcinella, Selection of Reference Signatures for Automatic Signature Verification, 5 th International Conference on Document Analysis and Recognition (ICDAR-5), Bangalore, India, Sept , 1999, pp [36] R. Kashi and W.L. Nelson, Signature Verification: Benefits of Multiple Tries, 8 th International Worshop on Frontiers in Handwriting Recognition (IWFHR-8), Niagara-on-the-Lae Ontario, Canada, Aug. 2002, pp [37] Mingming Ma, W.S. Wijesoma, Automatic online signature verification based on multiple models, IEEE/IAFE/INFORMS Conference on Computational Intelligence for Financial Engineering, 2000, pp [38] Mingming Ma, W.S. Wijesoma and Eric Sung, An automatic on-line signature verification system based on three models, Canadian Conference on Elect. and Comp. Eng., 2000, pp [39] C. de Oliveira, C.A A Kaestne Flávio Bortolozzi e Robert Sabourin, Generation of Signatures by Deformations, Proc. BSDIA, [40] B. Fang and Y.Y. Tang, Reduction of Feature Statistics Estimation Error for Small Training Sample Size in Off-Line Signature Verification, ICBA 2004, LNCS 3072, Springer-Verlag Berlin Heidelberg, D. Zhang and A.K. Jain (Eds.), 2004, pp [41] K. Huang and H. Yan, Off-line signature verification based on geometric feature extraction and neural networ classification, Pattern Recognition, Vol. 30, Issue 1, Jan. 1997, pp [42] B.Wirtz, Technical Evaluation of Biometric Systems, ACCU'98, Hong Kong, [43] G. Dimauro, S. Impedovo, G. Pirlo, G. Congedo, Signature Verification through a Dynamical Segmentation technique, Proc. 3th IWFHR, 1993, NY, USA, pp

SVC2004: First International Signature Verification Competition

SVC2004: First International Signature Verification Competition Dit-Yan Yeung 1, Hong Chang 1, Yimin Xiong 1, Susan George 2, Ramanujan Kashi 3, Takashi Matsumoto 4, and Gerhard Rigoll 5 1 Hong Kong University