A Deep Study on the Performance of the Spatial Density Distribution Method to Recognize Handwritten Signatures

Abstract

    A signature is a special identifier that confirms a person's identity and distinguishes him or her from others. The main goal of this paper is to present a deep study of the spatial density distribution method and the effect of a mass-based segmentation algorithm on its performance while it is being used to recognize handwritten signatures in an offline mode. The methodology of the algorithm is based on dividing the image of the signature into tiles that reflect the shape and geometry of the signature, and then extracting five spatial features from each of these tiles. Features include the mass of each tile, the relative mean, and the relative standard deviation for the vertical and horizontal projections of that tile. In the classification stage, four measurements of the Euclidean distance were used. While the accuracy rates for 4854 samples drawn from five different evaluated standard datasets ranged from 92.24% to 100%.