Spatial Analysis of Local Statistics for Handwritten Signature Recognition

 Handwritten signature is one of the most popular distinguishing biometric traits which can be used for secure personal authentication. Many challenges rise for handwritten signature recognition which include complexity of writing the signature (writing style, stroke pattern) also feature extraction and representation that give the best result. In this paper, a handwritten signature recognition system is proposed for static images. The system consists of three primary stages (preprocessing, feature extraction and recognition). In preprocessing stage, image processing methods are applied to remove the undesired noise and extract the signature region (ROI). After that, a new set of spatial-statistical features is determined from extracted ROI body, representing the density of the signature in each image block.  The set of introduced features is determined from the spatial domain after partitioning it into overlapped blocks. Then, the spatial-statistical features are determined from each block separately and assembled into one feature vector to represent the tested signature sample. The experimental results showed that the developed system could give recognition accuracy of around 99.81%; when tested on a dataset (SigCom2011) consisting of 612 signature images that belong to 102 persons using visual studio as programming environment.