Year: 2025
Type: article
Abstract: With the rapid development of deep learning technology and the extensive use of face recognition technology, the Transformer is gradually emerging in face recognition technology related applications. This paper systematically analyzes the evolution path of face recognition technology based on the Transformer architecture, focusing on the four technical perspectives of complex scene adaptation, occlusion robustness, lightweight design and low-scoring and high-noise image repair. In this paper, we conclude that Transformer brings a paradigm breakthrough in face recognition technology and demonstrates strong global sensing ability in complex scenes by virtue of its long-range dependent modeling advantage. With breakthrough potential at the level of cross-modal feature modeling and noise robustness, and an in-depth discussion on the innovative breakthroughs in cross-modal feature fusion and dynamic scene adaptation of the lightweight Transformer framework based on the self-attention mechanism,It also reveals the core bottlenecks such as strong data dependency and insufficient local texture sensitivity that the technique faces in practical applications. Ultimately, this paper validates the potential of Transformers in improving the robustness of face recognition through comprehensive evaluation and provides a theoretical framework and technology roadmap for scenario-based deployment of next-generation face recognition systems.
Author: Haiqing Jiang
Institution: Guangzhou City University of Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7081578726
FWCI:
Cites:
Cited by:
Related to: