IJDAR DOI 10.1007/s10032-012-0193-9 ORIGINAL PAPER Online signature verification based on signatures turning angle representation using longest common subsequence matching K. Barkoula · G. Economou · S. Fotopoulos Received: 5 December 2011 / Revised: 18 June 2012 / Accepted: 11 September 2012 © Springer-Verlag Berlin Heidelberg 2012 Abstract Online signature verification has been inten- sively investigated in several directions, such as the selected feature(s), similarity estimation and classification method. Local feature approaches combined with elastic distance metrics have the most successful performance so far. The Turning Angle Sequence (TAS) feature has not been exten- sively explored for signature verification, while the fusion of TASs of different scales, the Turning Angle Scale Space (TASS) is a new approach in this field. In this paper, we study the signatures TAS and TASS representations and their appli- cation to online signature verification. In the matching stage, a variation of the longest common sub-sequence matching technique has been employed. Experimental results using varying TAS(S) representation parameters on two publicly available signature databases, the SVC2004 and SUSIG, show the improved performance of the selected feature along with the chosen elastic distance measure on the equal error rate results of the online signature verification task. Keywords Online signature verification · Turning angle sequence · Least common subsequence matching 1 Introduction Signing is a widely used technique through centuries for person identification. However, the special intra-person vari- ability and its dependency on the acquisition process, time evolution and even on signer’s emotional state make signa- ture verification against forgeries a non-trivial task. K. Barkoula (B ) · G. Economou · S. Fotopoulos Department of Physics, University of Patras, Campus of Rion, Patras 26500, Greece e-mail: kbarkoula@gmail.com Signature verification, as presented in recent surveys [1–4], was initially a mainly offline technique where the sig- nature was captured as an image. Later on, with the develop- ment of solid sensors, the process evolved to online, in which case a signature is acquired using a digitizer and is repre- sented by dynamic information per point. The main phases of signature verification [1, 2] are data acquisition and pre- processing, feature extraction and classification. Data acqui- sition depends on the technology of the employed platform, and apart from trajectory coordinates and time, many other features, such as pressure, pen inclination and velocity, may be recorded [2], as can be found in online signatures data- bases [5–8]. Preprocessing techniques used with online signature’s fea- tures vary a great deal and may involve noise reduction, normalization, alignment, segmentation, translation, rota- tion and scaling invariant transformations. In [9], Gaussian filtering is used for noise reduction, while in [10], signature normalization is realized by employing Fourier Transform. Alignment based on the mass center and the principal axis of inertia is applied in [11]. Length-, time-based or pen up [12, 13] are some of the examined options for sig- nature segmentation. More advanced techniques, such as those based on signatures geometric characteristics [13–19], mainly use geometric extreme points. In [16–18], segmen- tation is carried out in two levels using a dynamic program- ming matching on vertical maximum in the first level and matching on vertical minimum in the second level. Fur- thermore, several criteria are used for identifying directly matched points. In [19], a Hidden Markov Model (HMM) is used for both segmentation and matching. In [20], sig- natures are initially aligned using Dynamic Time Warping (DTW) on the velocity feature. They are segmented using the relative angle between the slope of two consecutive seg- ments in the shape of the signature, splitting signatures’ 123