Offline signature verification based on pyramid histogram of oriented gradient features

Abstract

In this paper, we consider the problem of forgery and misuse of signatures that happen oftentimes. We propose a framework for offline signature verification using pyramid histogram of oriented gradient (PHOG) as a feature. The PHOG feature is extracted from a binary image of a signature of the same size that does not have much noise and therefore before it is extracted, a preprocessing image is performed. There are various parameters that may affect the extraction of PHOG characteristics such as number of bin, level, angular range, and amount of training data used. Using the best obtained parameters of PHOG descriptor and modified K-nearest neighbor (MKNN) classifier we get a 1.5% false rejection rate on the dataset of the Indonesian signature image that we obtain and 3% on the Persian signature image dataset, while false acceptance rate obtained on the Indonesian signature image dataset is 14.499% and 39.5% on the Persian signature image dataset.