International Journal of Engineering Trends and Technology (IJETT) – Volume 12 Number 6 - Jun 2014 Comprehensive Review of Offline Signature Verification #1 Rajinder Kaur, #2 Neha Pawar ,#3 Amit Chhabra #1 Student Mtech, #2 AssistanceProfessor, #3 Head of Deptt. Computer Science Department Swami Devi Dyal Institute of Engineering and Technlogy, Barwala,Panchkula Krukshetra university, krukshetra.Hyrayna-126102 India Abstract— In this paper we review the feasibility of combining continuous base classifiers for the purpose of off-line signature verification. The verification of signatures by taking the credentials of size and angle invariant for cheque system, there is probability of mismatch the characters in the signature of the user, here the Optical Character Recognition is used for automatic signature verification which can be helpful in many financial and business systems, but sometimes due to issues in colors of the character, font size, abstract or isolated character or symbol would be result into the mismatch of the signature, so our methodology for verification of signature is quite concrete to overcome such cases. NO. Off-line verification Online-verification 1. Verification depends upon analyzed image of person’s signature 2. Having a lot of noise included Information obtained slightly Fast verification process Verification depends upon capturing and analyzing the real time signature as person signing it Zero noise included 3. 4. Key words—DTI(Decision Tree Induction),NN(Neural Network),Sign(Signature),Diff(Differences),Pic(Picture). I. Fairly high degree of accuracy Table 1: Comparison of Off-line Verification and On-line Verification INTRODUCTION Signature verification is process of matching the signature in database with sample of signature given to be verified. Handwritten signature is widely accepted personal attributes for identity verification. The main area of research on signature verification is in the field of pattern recognition and image processing. It is also vastly used in the fields of finance, access control and securitythe signature verification is of two types:1) Offline verification 2)Online verification. In off-line verification system verification is done by off-line. The data is accessed by scanning individual handwritten signature. That scanned sign will be used for signature verification process. In on-line verification system verification is done on-line. The data accessed by touch screen, digitizer and stylus. This machines will make dynamic values such as location, pen pressure, co-ordinate values, speed of signature or time etc. In Off-line mode, it is complex to divide signature strokes due to highly stylish and unconventional writing styles. All of the complexity joined causes large intra-personal variation among sign .Biometrics offers many benefits over more commonly used authentication methods which provide a wide data material for study. More common such as photo ID cards or magnetic stripe cards can be lost, stolen, duplicated or even just left at home. Another common means of authentication is ISSN: 2231-5381 5. Information obtained by varying Faster than offline verification Very high degree of accuracy the use of passwords. There various things that can be go wrong with a password and in today’s fast paced world, people are enforced to remember a multiple of passwords for many things. Biometrics is fast to use, easy to use, accurate and authenticate. We need to use certain classifier that is going to play the role of decision making for verification. This phenomenon is termed as classification and is being done on certain criteria i.e. features that need to extracted from scan image of the sign to be verified. II. FEATURE EXTRACTION In this classification process, we need to look for certain criteria i.e. decide condition for matching signature sample with signature record in the database. This criteria is the feature linked to scanned sample of image of signature.A digital Image can be describe as two dimensional image as a finite set of discrete values, known as pic elements or pixels. Pixel ideals typically represent grey levels, color Intensity, heights, opacities etc. Image noise is the random variation of brightness or color information in images produced by the sensor and circuitry of a scanner or digital camera. Image noise is considered as an undesirable by-product of image capture. Image Processing is the analysis of a picusing predefined set algorithm let us know more detail information about the digital image.[3] http://www.ijettjournal.org Page 274 International Journal of Engineering Trends and Technology (IJETT) – Volume 12 Number 6 - Jun 2014 Due to the nature of static features and the adverse effect of background noise in scanning image. The offline systems made use of the certain characteristics of signature. The incorporation of intelligent image processing and feature extraction techniques, as well as robust classification models, are therefore key to the success of any off-line verification system. A.FORGERIES In the context of off-line signatures, forgeries may generally be categorised as eitherrandom, simple or skilled, in increasing order of quality. Furthermore, skilled forgeries may be subcategorised as either amateur or professional, as illustrated in Figure below . [4] Forgery Figure 2: Typical examples of (a) a genuine signature, as well as (b) professional skilled, (c)amateur skilled and (d) random forgeries. B.FEATURE EXTRACTED Random Simple Skilled Amateur Professional Figure1: Classification of forgeries 1) Random forgeries: Random forgeries encompass any arbitrary attempt at forging a signature, generally without prior knowledge of the owner’s name. This category of forgery may be random open strokes and easy to detect. The features of an image like pixel intensity, pixel no., pixel position, pixel area. Following feature could be extracted: 1) Pixel Intensity: The pixel density feature xPD ∈ [0, 1] has been used to great effect in such works asJustino et al. (2001) and Oliveira et al. (2005). Since the pixel density of a signature segment is directly linked to stroke width, it is also commonly referred to as apparentpen pressure. The pixel density of an image J is obtained by computing the ratio of pen stroke pixels to total image pixels, or (1) 2)Simple forgeries: Simple forgeries are used when the forger’s knowledge is restricted to the name of the signature’s owner. Due to the arbitrary nature of signature design, simple forgeries may in some cases bear an alarming resemblance to the writer’s genuine signature. In such cases, more sophisticated systems, able of detecting subtle stylistic diff., are required in order to distinguish between genuine signatures and forgeries of this type.[6],[9],[2] 3)Skilled forgeries: In some instances, the forger is not only familiar with the writer’s name, but also has access to samples of genuine signatures. Given ample time to practice signature reproduction, he is able to produce so-called skilled forgeries. The vast majority of skilled forgeries may be categorized as amateur, as this type of forgery may be produced by any given individual. In contrast, to produce a professional skilled forgery, the forger typically requires a certain amount of knowledge regarding forensic document analysis. The duplication of the original signature, which means is that the person recognizes that how the original signature exactly looks like .So we can say that skilled signatures are the most tuff to detect. ISSN: 2231-5381 2)Gravity Center: The gravity center distance feature xGCD ∈ [0, 1], as described in Justino et al. (2005), is Obtainedin two phases. Firstly, the coordinates of the image centroid (¯x, ¯y) are computed using (2) Thereafter, the gravity center distance associated with J, as illustrated in Figure 3.4, can be computed as follows, http://www.ijettjournal.org (3) Page 275 International Journal of Engineering Trends and Technology (IJETT) – Volume 12 Number 6 - Jun 2014 [4] Figure 4: The perceptron can learn by adjusting theweights to approach the desired output. Figure 3: Computation of the gravity centre distance feature. 3)Image Area: It is defined as no of pixels in width and no of pixels in height of an image. let us assume that X is no of pixels in width of Image and Y is no of pixels in Height of an image, or Image Area = X x Y (4) III CLASSIFIERS Classification is a method by which labels or class identifiers are attached to the pixels making up a remotely sensed image on the basis of their characteristics. These characteristics are generally measurements of their spectral response in different wavebands. They may also include other attributes (e.g., texture) or temporal signatures. In order to combine the classifier ensemble constructed from these base classifiers, several classifier combination strategies is considered. Neural Network algorithm of self learning is investigated. The efficiency of this method is greatly increased by the sigmoidal score normalization function utilized in this study. The other decision classifier Techniques Are Also Investigated, Namely The Neural Networks Classifiers and Decision Tree Induction Classifiers The popular classifier ensemble combination strategy. In constructing a classifier ensemble, any number of base classifiers may be utilized, regardless of their feature extraction or signature modeling techniques. The optimal ensemble composition is determined experimentally. X1<=n1 X4<=n4 Neural networks do not follow a set of instructions, provided for them by the author, but they learn as they go case by case. Neural networks are highly devoted when trained using a large amount of data. They are used in applications where security is highly valued. In this research we used Multi-Layer Perceptron MLPs neural network. The structure is depends of the neural network on all the layer providing significant feedback, where all the nodes are connected to all other nodes in the next layer and so on, but these nodes do not have any connections with the layer below it. Then, it was being modified to function as a back propagation neural network, using the BP algorithm.[7] X3<=n31 Y=3 X3<=n32 Y=1 Y=3 X2<=n2 Y=2 Y=1 A NEURAL NETWORKS CLASSIFIERS ISSN: 2231-5381 B.DECISION TREE INDUCTION CLASSIFIERS An alternative approach is to use a multistage or sequential hierarchical decision scheme. The basic idea involved in any multistage access is to break up a difficult decision into a set of several simple decisions, hoping the final solution achieved in this way would simulate the expected desired solution. Hierarchical classifiers are special type of multistage classifier that allows rejection of class labels at intervening stages. Y=1 X3<=n3 Y=3 Figure 5: A DTI classification tree for a five dimensional features and three classes.[4] Classification trees present an effective implementation of such hierarchical classifiers. Classification trees have become progressively important due to their imaginary simplicity and computational efficiency. A decision tree classifier has a simplest form which can be easily compressed then stored in storage device and that can accurately and sufficiently classifies new data. [1],[5]Decision tree classifiers can act automatic feature selection and difficulty reduction, and their tree structure gives easily understand and intelligible information about the predictive or generalization ability of the classification. To construct a classification tree by http://www.ijettjournal.org Page 276 International Journal of Engineering Trends and Technology (IJETT) – Volume 12 Number 6 - Jun 2014 heuristic approach, it is assumed that a data set expressing of feature vectors and their similar class labels are available. The features are recognized based on problem specific knowledge. The decision tree is then composed by repeat partitioning a data set into further more parts, more homogenous parts on the basis of a set of tests applied to one or more attribute values at each branch in the DTI tree. This procedure has three steps: splitting nodes, determining which nodes are terminal nodes, and assigning class label to terminal nodes.[10] In general, a pattern recognition system is constructed by utilizing one or more feature extraction techniques, in conjunction with a single classification technique. The use of several feature extraction techniques is recommended, as this ensures greater separation of different pattern classes in the feature space. Given a set of two or more classifiers, referred to as a classifier ensemble, it is logical to expect an improvement in performance when combining the separate efforts of each into a single classifier, referred to as a combined classifier. The combination process is performed either on score level or decision level. IV CONCLUSION We here analysed and conferred that an off-line signature verification system which is based on simple geometric features. The geometrical features were chosen carefully so as to sufficiently distinguish signatures from different people. The study show that using a localized threshold out performs a global threshold by a vast margin. The localized thresholds ability to correctly classify a signature is up by more than the global threshold. The false acceptance rate is down when using localized thresholds. Improved performance may be obtained by using a higher number of features as well as incorporating the capture of dynamic data as it is this dynamic data that uniquely identifies a person. The system can be improved using the concepts of Neural Networks and DTI Algorithm which hold a lot of promise of being able to organize a system that performs with a very high level of accuracy. V FUTURE SCOPE likely be achieved by either considering a fundamentally different set of features and/or signature modeling techniques. REFERENCES [1]Ashwini Pansare,Shalini Bhatia,Handwritten Signature Verification using Neural Network,International Journal of Apllied Information Systems,Volume 1-No,2012, pp. 44-49. [2Jackowski, k. and wozniak, m. (2009). adaptive splitting and selection method of classifier ensemble building, pp. 525–532. lecture notes in computer science. springer. [3]odeh,s.m.,khalil,m., off-line signature verification and recognition: neural network approach , innovations in intelligent systems and applications (inista), 2011 international symposium on ,2011,pp 34-38, isbn:978-1-61284-919-5 [4] Lakshmi, k.v. electron. & commun. Dept., echelon inst. Of technol., Faridabad, india nayak, s., offline signature verification using neural networks, advance computing conference(iacc), 2013 ieee 3rd international [5] Ferrer, m.a. ; offkline signature verification using local patterns morales, a. ; vargas, j.f., telecommunications (conatel), 2011 2nd national conference on ,pp1-6, isbn:978-1-4577-1046-9 [6]M. Ferrer, J. Alonso, and C. Travieso, “Offline Geometric Parameters for Automatic Signature Verification Using Fixed-Point Arithmetic”, IEEE transactions on pattern analysis and machine intelligence, vol. 27,no. 6, June 2005. [7]S. rasoul safavian and david landgrebe ,A survey of decision tree classifier methodology ieee trans. systems, man, & cybernetics,2010, [8] Jacques philip swanepoel, bschons (stell),off-line signature verification using classifier ensembles and flexible grid features, financial assistance of the national research foundation (nrf),2009 [9] Coetzer, j., herbst, b. and du preez, j. (2004). off-line signature verification using the discrete radon transform and a hidden markov model. eurasip journal on appliedsignal processing, vol. 4, pp. 559–571. [10Jcoetzer, j., herbst, b. and du preez, j. (2006). off-line signature verification: a comparison between human and machine performance. international workshop on frontiers in handwriting recognition, vol. 10, pp. 481–485. [11]Dos santos, e., sabourin, r. and maupin, p. (2008). a dynamic overproduce-and-choose strategy for the selection of classifier ensembles. pattern recognition, vol. 41, pp. 2993–3009 During the course of this reviewing, certain concepts were encountered that warrant further investigation, as they are deemed potentially beneficial to the systems developed in thisstudy. These concepts are not included in this work, either due to being considered outside the scope of this study, or simply due to time constraints.These topic were Adaptive grid segmentation and Conditionally independent classifier ensembles. The flexible segmentation grids could be developed in this study, however, are subject to the requirementthat each grid cell boundary is dilated, if possible, to exactly the same extent. Definitely warrants further investigation to attempt the design and implementationof such a conditionally independent classifier ensemble. The requirement of independent classifier decisions may most ISSN: 2231-5381 http://www.ijettjournal.org Page 277