Deep-HFID: Deep Neural Network Based Hands and Face bio-metric Identification System Using Metric Learning
Zahra Parvin Ashtyani, Sara Ghanbari, Navid Zare, Mehdi Tale Masouleh, Ahmad Kalhor · 2022 10th RSI International Conference on Robotics and Mechatronics (ICRoM) · 2022
This paper presents a device for person identification based on hand bio-metrics. The main objective of a person recognition device is to perform accurate identification at a reasonable production cost. The so-called Deep-HFID device is designed to meet these objectives as well as to provide users with ease of use. For the purpose of identifying individuals, this device uses hand and face bio-metrics. In terms of dimensions, this device has been designed to facilitate the use of a variety of individuals. For ease of deployment, the Deep-HFID system relies on a server for most of its processes. The routine tasks, like showing the graphical user interface are handled by a Raspberry Pi board. A dataset from the palm side of both hands from 110 individuals has been created. Using transfer learning, the six pre-trained state-of-the-art neural network architecture was fine-tuned and calibrated based on the F1 score and precision criteria. An augmented dataset containing samples from the palm side of 30 individuals with between fingers angles variation is collected. By applying the main and augmented dataset and using the center-based separation index as a complexity measure algorithm to rank pre-trained deep neural networks, a compression on the last layer of the best performed neural network, EfficientNetV2B1, is done which results in a more robust system without reducement in accuracy.