ABSTRACT
This project presents a robust deep learning-based approach for real-time face classification and emotion detection using a combination of convolutional neural networks (CNNs) and the DeepFace library. The primary objective is to accurately identify individuals from facial images and simultaneously detect their emotional states in live video streams. A custom dataset comprising labeled face images is used to train the classification model. The dataset is organized into separate directories for each identity class, enabling efficient supervised learning. For face classification, a CNN model is developed using the Keras Sequential API. The architecture includes multiple convolutional and pooling layers followed by dense layers and dropout regularization. All images are resized to 224x224 pixels and normalized to ensure optimal model performance. One-hot encoding is applied to the class labels, and the dataset is divided into training and testing sets with a 70:30 split. The model is compiled with RMSprop optimizer and categorical cross-entropy loss and trained over 30 epochs.
To ensure reliable face detection, the Haar cascade classifier is employed, which identifies faces from video frames captured in real-time through a webcam. Once a face is detected, the model predicts the person's identity. In parallel, the DeepFace library is utilized to analyze the same face region for emotion classification. DeepFace supports multiple emotions like happy, sad, angry, surprise, and more, making it well-suited for emotion-aware applications. The system displays the predicted identity and emotion on the video frame, providing an interactive visual output. Evaluation metrics such as accuracy, precision, recall, and F1-score confirm the effectiveness of the trained CNN model. Additionally, training and validation performance are visualized using accuracy and loss curves. The trained model is saved and deployed for real-time inference.
This project demonstrates a successful integration of computer vision and deep learning techniques for dual-purpose recognition tasks. It has promising applications in surveillance, human-computer interaction, sentiment-aware systems, and biometric authentication, offering a scalable solution for real-world scenarios.


OBJECTIVES
1. To develop a real-time facial recognition system using a Convolutional Neural Network (CNN) that can accurately classify individuals based on facial images captured from a webcam.
2. To integrate emotion detection functionality using the DeepFace library, enabling the system to identify dominant emotions such as happy, sad, angry, neutral, and surprised from live video input.
3. To implement an efficient face detection mechanism using Haar cascade classifiers for accurately locating facial regions in each frame of the video stream before classification and emotion analysis.
4. To train and evaluate a deep learning model on a custom dataset of face images, using appropriate metrics such as accuracy, precision, recall, and F1-score to validate the performance of the face classification system.
5. To design a unified and interactive real-time system that displays both the predicted identity and emotional state of the detected person on-screen, suitable for applications in surveillance, human-computer interaction, and emotion-aware environments.