Histogram of oriented gradient (HOG) for off-line handwritten signature authentication

Abstract

Security is needed as it could secure the routine activities particularly when it comes to document-related works. The signature authentication system is being used widely in this era of technology to authenticate an individual’s identity. Yet, several problems such as the increasing number of forge signature still occur, which induced a need for an efficient signature authentication system. The human eyes could not differentiate between the genuine and forge signature, and the traditional method to authenticate a signature in financial institutions mainly is very time-consuming. Thus, an off-line handwritten signature authentication using a technique of Histogram of Oriented Gradient (HOG) is proposed. HOG is a gradient-based feature extraction method. Initially, the offline handwritten signature images will go through the pre-processing and image enhancement processes such as binarization, thinning, noise removal, and resizing. Next, the shape feature is extracted using the HOG which produced the feature vector values. Different images will have different feature vector values. These values are then be used to measure the similarity between the genuine and tested signatures using Euclidean distance. Three threshold values are used for the performance analysis of the HOG signature authentication which are 15, 20, and 30. From the analysis conducted, it is found out that 15 and 20 are the best threshold values for HOG which returned good performance of 87.5% for Overall Accuracy (OA). It is expected that the outcome of this study may assist the user in the off-line handwritten signature authentication in the future.