Off-line Handwritten Signature Verification Using Contourlet Transform and Co-occurrence Matrix Assia Hamadene, Youcef Chibani and Hassiba Nemmour Speech Communication and Signal Processing Laboratory, Faculty of Electronics and Computer Science University of Science and Technology Houari Boumediene (USTHB), 32, El Alia, Bab Ezzouar, 16111, Algiers, Algeria ahamadene@usthb.dz , ychibani@usthb.dz , hnemmour@usthb.dz. Abstract—We address, in this work, a new feature generation method for two different approaches of off-line handwritten signature verification (HSV), writer-dependent and writer- independent HSV. The proposed method uses conjointly the contourlet transform and the co-occurence matrix. The contourlet transform allows capturing contour segment directions of the handwritten signature, while the co- occurrence matrix allows describing the number of directions. Experiments are conducted on the well known CEDAR dataset and the classification through the support vector machines (SVM). The obtained results show the effective use of the Contourlet transform for handwritten signature verification comparatively to the state of the art. Keywords-off-line Handwritten Signature Verification; Contourlet transform; SVM; co-occurrence matrix; writer dependent; writer independent. I. INTRODUCTION The handwritten signature verification (HSV) is a discipline which aims to validate the identity of writers according to the handwriting styles [1]. It is one of the most widely used for being simple, inexpensive and acceptable from society. However, it also represents one of the easiest breakable security systems compared to the physiological biometric ones, since signatures can easily be imitated. Hence, the signature verification is still an open problem because a signature is judged to be genuine or a forgery only on the basis of a few reference specimens [2][3]. Furthermore, a same writer can sign differently depending on his or her state of emotion. The design of a Handwritten Signature Verification System (HSVS) depends on the acquisition mode of the signature. The first mode, called on-line or dynamic acquisition, allows capturing some dynamic characteristics of the written style such as velocity, pressure and acceleration. The second mode, called off-line or static acquisition, allows generating an image, which represents a more difficult task due to the disappearance of dynamic features. However, this mode is still the most applicable in daily cases. Two different approaches can be adopted for offline signatures verification [4]. The usual approach is the writer dependent HSV, where models for genuine and forgery signatures are constructed for each writer. Then, the questioned signature sample of a writer is compared to its own model. The disadvantage of this approach is the need to generate a model for each new writer to be verified. The second approach called writer-independent HSV is used by forensic experts [4]. This approach is considered as the most practical cases, since it is not necessary to generate a model for each writer in order to verify its signature. In this case, a general model is built from some writers chosen randomly. However, the writer-independent HSV constitutes a more difficult task because of the important morphological variability inter-writers. Generally, a HSV system is composed of three main stages: data acquisition and preprocessing, feature generation and classification. During the classification stage, personal features generated from an acquired signature are compared against features of the reference signatures stored in the database in order to judge its authenticity [3]. Hence, the feature generation stage plays an important role for the robustness of a HSVS. Various methods have been developed for generating features from the signature image, which can be grouped into two categories: direct methods and transform methods. Direct methods allow generating features directly from image pixels such as grid-based information, pixel density, gray-level intensity, texture… etc. In contrast, transform methods need a transformation of the image into another domain in which features could be generated. Fourier, Wavelet, Radon transforms are the most popular methods for generating features [3][5]. The main drawback of these methods is that they don’t allow capturing contours contained into an image. Hence, a sophisticated transform has been proposed more recently namely the contourlet transform (CT) [6]. The main advantage of the CT is the ability to capture significant information about an object. Furthermore, it offers a flexible multiresolution, local and directional image expansion [6]. These properties are interesting to exploit more specifically for the handwritten signature verification since the signature contains often special characters and flourishes [7]. The CT has successfully been used for many applications such as handwritten signature verification [1][2], vehicle recognition [8], noise reduction of biomedical images [9], face recognition [10] and image retrieval [11][12]. 2012 International Conference on Frontiers in Handwriting Recognition 978-0-7695-4774-9/12 $26.00 ' 2012 IEEE DOI 10.1109/ICFHR.2012.245 343