A Benchmark Database for the Comparison of Face Morphing Detection Methods
Le-Bing Zhang, Juan‐Juan Cai, Fei Peng, Min Long · 2021 International Conference on Electronic Information Technology and Smart Agriculture (ICEITSA) · 2021
Aim to provide a public and fair comparison platform for evaluating the performance of face morphing detection, a benchmark database HNU-FM is built in this paper. It contains 100 genuine subjects. All morphed face images are automatically generated following the guidance of complete morphing and splicing morphing, and different morphing factors are taken into account. The morphed images are all validated by Face++. Different from the existing databases for face morphing detection, it contains balanced samples between the real face images and the morphed face images, and four evaluation protocols are developed for evaluating different application scenarios. With Utrecht_M and HNU-FM, experiments were conducted on some existing face morphing detection methods for performance comparison. Experimental results and analysis show that the proposed HNU-FM database can further effectively evaluate the stability of various face morphing detection methods on the variation of morphing factors, which are impossible to be evaluated in Utrecht_M. It also shows that HNU-FM has great potential to be a public evaluation benchmark database for face morphing detection.