Morphing Attack Detection: A Fusion Approach
Siri Lorenz, Ulrich Scherhag, Christian Rathgeb, Christoph Busch · 2021 IEEE 24th International Conference on Information Fusion (FUSION) · 2021
Face morphing attacks pose a serious threat to existing face recognition systems. As a number of studies have shown, existing face recognition systems and human experts can be fooled by morphed facial images. Based on these findings various approaches to morphing attack detection have been published. Automated morphing attack detection is still a young branch of research with many recent publications.Using features extracted by different feature extractors we develop a score-level based fusion approach. The scores are gen- erated by different classifiers with optimised hyperparameters. We use different approaches to determine the weights for the sum-rule: grid-search and random forests scoring function as well as normalised scores.We notice that a weighted score-level fusion can achieve improved results. Moreover, we observe that weights determined by grid-search might lead to better results when using fewer scores compared to those obtained by random forest while the former is more time consuming. However, both random forest and grid-search weights can significantly improve the morphing attack detection performance.