In this work, a new approach for off-line signature recognition and verification is presented and described. A subset of the line, concave and convex family of curvature features is used to represent the signatures. Two major constraints are applied to the feature extraction algorithm in order to model the two step transitional probabilities of the signature pixels. Segmentation of the signature trace is enabled using a window which is centred upon the centre of mass of the thinned image. Partitioning of the image leads to a multidimensional feature vector which provides useful spatial details of the acquired handwritten image. The classification protocol followed in this work relies on a hard margin support vector machine. Our method was applied to two databases, the first taken from the literature while the second created by the authors. In order to provide comparable results for the first stage signature verification system, we have applied an already published feature extraction method while keeping the same classification protocol. Primary evaluation schemes on both corpuses provide very encouraging verification results for the Average Error.