This project implements a facial recognition system using a deep learning approach with GoogleNet. It includes training the model with an augmented image dataset and testing it via webcam input. The system captures images, resizes them, classifies them using a pre-trained network, and displays the results.
- We utilized the GoogleNet architecture to train a model for facial recognition. The dataset is split into training and validation sets, with labels sourced from folder names.
- Layers 142 and 144 of GoogleNet are replaced with a fully connected layer and a classification layer, customized for facial recognition based on the number of facial classes.
- Image augmentation techniques such as random reflection, translation, and scaling are applied to increase model robustness and improve generalization.
- The model is trained using Stochastic Gradient Descent with Momentum (SGDM), and the training progress is visualized using MATLAB's tools.
- The trained model is tested using real-time images captured from a webcam.
- Each captured image is resized to match GoogleNet’s input size (224x224) and classified using the trained model.
- The image is then displayed alongside the predicted label and confidence score.
- Dataset Augmentation: Enhances model generalization by introducing variations in the training data.
- Custom Layer Integration: GoogleNet’s feature extraction layers are fine-tuned for facial recognition.
- Real-Time Testing: Captures images from a webcam and performs classification on-the-fly, displaying the result.
- MATLAB: For training, augmenting images, and performing real-time classification.
- GoogleNet: A pre-trained Convolutional Neural Network (CNN) modified for facial feature extraction and classification.
This project provides a robust framework for facial recognition, which can be further adapted for other image classification tasks.
Some of the images were collected from Google Images.