University of Twente Student Theses


Patch-Based Morphing Attack Detection

Jiang, Yuling (2023) Patch-Based Morphing Attack Detection.

[img] PDF
Abstract:The morphing attack poses a significant threat to face recognition systems, as it undermines the unique link between identity and identification documents. Therefore, the need for morphing attack detection is imperative. In this paper, we proposed a novel patch-based morphing attacks detection approach. The facial regions from the images were cropped and divided into 30 patches. This methodology facilitates a straightforward expansion of the dataset size. We conducted a comprehensive analysis comparing different combinations of feature extraction networks and score fusion mechanisms. The findings demonstrate that the utilization of Se_Resnet50 as the feature extractor, combined with either the average or machine learning score fusion method, produces satisfactory results during the test phase. In particular, the D-EER for intra-dataset tests is 0%, and the highest D-EER observed for cross-dataset tests is merely 12.1%. However, when conducting cross-dataset testing, morphs generated with STYLEGAN2 exhibit an exception to this trend. Subsequently, extensive experiments with the optimal combination were conducted to investigate the influence of various training settings on the outcomes. The findings unveiled that when subjected to images generated by STYLEGAN2 from distinct datasets, a model exclusively trained on STYLEGAN2-generated images exhibited enhanced capabilities in generalization. Furthermore, there are certain similarities in the artifacts observed in both landmark-based and GAN-based morphed images.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science MSc (60300)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page