Enhanced face morphing attack detection using error-level analysis and efficient selective kernel network

Abstract

Face morphing attack detection is a research hotspot in the field of biometrics. However, existing methods cannot balance accuracy and complexity well due to their inability to effectively capture crucial feature distinctions. To solve this problem, this paper proposes a detection method based on error-level analysis and an efficient selective kernel network. Specifically, the proposed method performs error-level analysis on the R, G, and B color channels to accurately capture crucial feature differences and enhance face detection accuracy. Meanwhile, an efficient selective kernel network is designed with further improvements and optimizations. This network can adaptively adjust the size of the receptive field, thereby providing higher classification accuracy without significantly increasing the number of parameters. Besides, the shallow feature enhancement module and feature fusion module are designed to improve the model's detection performance. Finally, the proposed method is evaluated on standard databases and compared to existing detection methods. The results demonstrate that the proposed method achieves superior detection performance to existing methods.