OffSig-sinGAN: A deep learning-based image augmentation model for offline signature verification / Muhammad Muzaffar Hameed

Muhammad Muzaffar , Hameed (2023) OffSig-sinGAN: A deep learning-based image augmentation model for offline signature verification / Muhammad Muzaffar Hameed. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (203Kb)
	PDF (Thesis PhD) Restricted to Repository staff only until 31 December 2024. Download (2086Kb)

Abstract

Offline signature verification (OfSV) is essential in preventing the falsification of documents. In a real-world scenario, forensic handwriting experts verify offline signatures by visually comparing the suspected with the reference. However, the authenticity of the questioned signature is estimated by evaluating the characteristic features of signatures, their similarities and differences within the suspected and reference samples. Forensic handwriting experts can assess the probability that the evidence under investigation indicates that the questioned signature is an authentic signature used by the reference signer, or the questioned signature is the product of a forgery process with the utilisation of computational methods, such as OfSV systems. Deep learning (DL)-based OfSV systems have recently attained state-of-the-art results in offline signature verification. However, these systems require a large number of signature images to achieve acceptable performance. Nonetheless, only a limited number of signature samples are readily available to train these models in real-world scenarios. Several researchers have proposed models to augment new signature images by applying various transformations. Some have used human neuromotor and cognitive-inspired augmentation models to address the demand for more signature samples. However, such approaches fail to create a considerable number of signatures and to bring any new visual features to improve the network learning ability. Augmenting a sufficient number of signatures with variations is still challenging. Therefore, this study proposed OffSig-SinGAN, which is a deep learning-based image augmentation model to address the limited number of signature problems on offline signature verification. The proposed model can augment several high-quality signature images with diversity from a single signature image only. The quality of augmented signature images is assessed using four metrics which are pixel-by-pixel difference, peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and frechet inception distance (FID). Various experiments were conducted to evaluate the proposed image augmentation model's performance on selected DL-based OfSV systems and to prove whether it helped to improve the verification accuracy rate. Experiment results showed that the proposed image augmentation model (OffSig-SinGAN) performed 2.93% better on the GPDSsyntheticSignature dataset than other augmentation methods. The improved verification accuracy rate of the selected DL-based OfSV system proved the effectiveness of the proposed augmentation model.

Actions (For repository staff only : Login required)