| Summary: | This project is a study on how Convolutional Neural Networks (CNN) can be utilised to recognise faces occluded with accessories, specifically face masks. The widespread use of face masks, particularly highlighted by the COVID-19 pandemic, has significantly impaired the effectiveness of traditional face recognition systems. To overcome this limitation, this study proposes an approach that trains a CNN model on a dataset of unoccluded faces, aiming to maintain high recognition accuracy even when faces are partially obscured.
The project encompasses several phases, including the creation of a comprehensive face database, pre-processing of face images to facilitate optimal model training, and the development of a software-based recognition system implemented in Python. Special emphasis is placed on experimenting with data augmentation and hyperparameter tuning to improve model robustness.
A key component of the project is the development of a web application, designed to demonstrate the model's capabilities. This application, encapsulated within Docker containers for ease of deployment, allows users to interact with the system by capturing and uploading images of their face for recognition, thereby showcasing the practical applications of the developed model.
Preliminary results indicate that the CNN model achieves commendable accuracy in recognizing unoccluded faces. However, recognizing faces with accessories poses a greater challenge, highlighting the need for further model refinement. The project's findings contribute valuable insights into the capabilities and limitations of using CNNs for face recognition in the context of faces occluded with surgical masks, offering a foundation for future research and development in this critical area of biometric authentication.
|
|---|