Facial Recognition and Face Mask Detection Using Machine Learning Techniques
Saved in:
| Title: | Facial Recognition and Face Mask Detection Using Machine Learning Techniques |
|---|---|
| Authors: | Boulos, Mira M. |
| Summary: | Facial recognition, as a biometric system, is a crucial tool for the identification procedures. When using facial recognition, an individual's identity is identified using their unique facial features. Biometric authentication system helps in identifying individuals using their physiological and behavioral features. Physiological biometrics utilize human features such as faces, irises, and fingerprints. In contrast, behavioral biometric rely on features that humans do, such as voice and handwritings. Facial recognition has been widely used for security and other law enforcement purposes. However, since COVID-19 pandemic, many people around the world had to wear face masks. This thesis introduces a neural network system, which can be trained to identify people’s facial features while half of their faces are covered by face masks. The Convolutional Neural Network (CNN) model using transfer learning technique has achieved remarkable accuracy even the original dataset is very limited. One large Face mask detection dataset was first used to train the model, while the original much smaller Face mask detector dataset was used to adapt and finetune this model that was previously generated. During the training and testing phases, network structures, and various parameters were adjusted to achieve the best accuracy results for the actual small dataset. Our adapted model was able to achieve a 97.1% accuracy. |
| URL: | https://digitalcommons.montclair.edu/etd/728 |
| Database: | OpenDissertations |
| Abstract: | Facial recognition, as a biometric system, is a crucial tool for the identification procedures. When using facial recognition, an individual's identity is identified using their unique facial features. Biometric authentication system helps in identifying individuals using their physiological and behavioral features. Physiological biometrics utilize human features such as faces, irises, and fingerprints. In contrast, behavioral biometric rely on features that humans do, such as voice and handwritings. Facial recognition has been widely used for security and other law enforcement purposes. However, since COVID-19 pandemic, many people around the world had to wear face masks. This thesis introduces a neural network system, which can be trained to identify people’s facial features while half of their faces are covered by face masks. The Convolutional Neural Network (CNN) model using transfer learning technique has achieved remarkable accuracy even the original dataset is very limited. One large Face mask detection dataset was first used to train the model, while the original much smaller Face mask detector dataset was used to adapt and finetune this model that was previously generated. During the training and testing phases, network structures, and various parameters were adjusted to achieve the best accuracy results for the actual small dataset. Our adapted model was able to achieve a 97.1% accuracy. |
|---|
