Abstract: The objective of this project is to design an energy-efficient face authentication system from face recognition by capitalizing on advancements in deep learning and hardware optimization. With the use of model compression techniques like pruning and quantization, and specialized hardware like Google Coral Edge TPUs, our system achieves tremendous energy savings without affecting accuracy. Our system uses a multi-step procedure beginning with face detection using a light CNN. The detected face is then passed through an energy-conserving compressed face recognition model via methods like knowledge distillation and weight sharing. Our system is further optimized through the utilization of the Google Coral Edge TPU, which is just amazing with minimal energy consumption. We contrast the operation of our system on multiple datasets, such as LFW and IJB-A, and demonstrate that it computes at state-of-the-art accuracy on using less energy than other face recognition systems.

Keywords: Face Recognition, Energy Efficiency, Deep Learning, Model Compression, Hardware Optimization, Edge Computing, Authentication, Biometrics, Security, Google Coral Edge TPUI.

| DOI: 10.17148/IARJSET.2025.125389