Developing Deep Learning Based Facial Recognition Technique | ||
| Journal of Advanced Engineering Trends | ||
| Volume 44, Issue 1, January 2025, Pages 214-220 PDF (465.21 K) | ||
| Document Type: Original Article | ||
| DOI: 10.21608/jaet.2024.322195.1344 | ||
| Authors | ||
| Hossam Mahmoud Elian1; Gamal M. Dousoky* 1; Ali Hafez2 | ||
| 1Electrical Engineering Dep., Faculty of Engineering, Minia University, Minia, Egypt | ||
| 2National Research Institute Of Astronomy and Geophysics, Giza, Egypt | ||
| Abstract | ||
| Identity verification is becoming more and more crucial in a variety of real-world applications, including identity checks at airports, apartment door locks, and cell phones. This process requires a methodology that is fast, precise, scalable to accommodate additional users, and adaptable to variations in face angle, brightness, and other variables. In order to address the aforementioned difficulties, we provide four facial recognition techniques in this work. First, the CNN architecture was presented. Then, We developed the CNN Decoder for face encoding to improve model accuracy and overcome the difficulty of retraining the model when adding new people. Next, we presented two capsule network topologies to address face angle-related problems. The COMSATS Face Dataset is the dataset that we used in our study for testing, training, and assessment. According to an experiment, CNN recognizes faces with 93% accuracy, the decoder with 99% accuracy, the CapsNet with CNN 81% accuracy, and the CapsNet with VGG-19 with 99% accuracy. The latter is thought to yield the greatest results when it comes to distinguishing faces from various viewing angles. | ||
| Keywords | ||
| Face Recognition; Deep Learning; Auto-Encoder; Capsule Network | ||
|
Statistics Article View: 55 PDF Download: 62 |
||
