Scribd
Upload
8 views7 pages

MiniProject DL

Uploaded by

Anushka Shingade
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views7 pages

MiniProject DL

Uploaded by

Anushka Shingade
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

SCTR’s Pune Institute of Computer Technology, Pune

A PROJECT REPORT ON Human Face Recognition.

SUBMITTED BY

Class : BE 4 (S4)

Name : Manasi Mahadev Raut


Roll No. : 41468

Under the guidance of


Prof. A.D.Bundele

DEPARTMENT OF COMPUTER ENGINEERING


Title: Human Face Recognition

Problem Statement: Human Face Recognition.

Learning Objectives:
1. Understand the fundamentals of deep learning for face recognition.
2. Implement a convolutional neural network (CNN) for facial feature extraction.
3. Train and evaluate a deep learning model on a face recognition dataset.
4. Analyze accuracy, efficiency, and real-time applicability of the model.

Learning Outcomes:
1. Ability to implement deep learning-based face recognition models.
2. Understanding of CNNs for facial feature extraction and classification.
3. Proficiency in training and evaluating face recognition models using datasets.
4. Capability to analyze model performance, accuracy, and real-world applicability.

Theory:
1. Introduction to Face Recognition
• Face recognition is a biometric technology used for identifying or verifying
individuals based on facial features.
• It has applications in security, surveillance, authentication, and human-
computer interaction.
• Deep learning, particularly Convolutional Neural Networks (CNNs), has
significantly improved face recognition accuracy.

2. Fundamentals of Deep Learning for Face Recognition


• Neural Networks: Computational models that mimic the human brain’s ability
to recognize patterns.
• Deep Learning: A subset of machine learning that uses multiple layers of
artificial neurons for feature extraction.
• Supervised Learning: The most common approach, where the model is trained
on labelled face images.
3. Convolution Neural Networks (CNNs) for Face Recognition
• CNNs are widely used for image processing due to their ability to
automatically extract hierarchical features.
• Key Layers in CNNs:
o Convolutional Layer: Extracts features using filters.

4. Dataset and Preprocessing


• Common Face Recognition Datasets:
o Labelled Faces in the Wild (LFW) – A benchmark dataset for face
verification. o CASIA-WebFace – Large-scale dataset for training
deep learning models.
• Preprocessing Techniques:
o Face alignment and cropping.

5. Model Training and Evaluation


• The dataset is divided into training, validation, and test sets.
• Performance is evaluated using metrics like:
o Accuracy: Measures correct predictions.
o Precision C Recall: Assess class-wise performance. o ROC Curve C
AUC: Evaluate classification confidence.

6. Challenges in Face Recognition


• Variations in lighting, pose, and facial expressions.
• Occlusions due to accessories (glasses, masks). • Data privacy
and ethical concerns in facial recognition applications.
System Architecture:
The system architecture for deep learning-based face recognition consists of multiple stages,
including image acquisition, preprocessing, feature extraction, and classification. The
architecture is designed to ensure accurate and efficient face identification or verification.

1. Components of the Architecture

o Image Acquisition o Captures facial images using cameras or loads images


from a dataset.
o Can include real-time video input for dynamic face recognition.
o Preprocessing Module o Detects and aligns faces using algorithms like
MTCNN
o Crops and normalizes facial images to a standard size. o Applies data
augmentation techniques (rotation, scaling, flipping) to enhance model
robustness. o Feature Extraction using CNNs o A Convolutional Neural
Network (CNN) extracts facial features by learning hierarchical
representations. o The extracted features are mapped into an embedding space
where similar faces have closer distances. o Pre-trained architectures like
FaceNet, VGGFace, and ResNet are commonly used for feature extraction. o
Classifications Recognition o The extracted embeddings are compared using
similarity measures like cosine similarity or Euclidean distance.
o For classification-based models, a fully connected layer with a softmax
activation function is used.
o The system determines the identity of the person based on the nearest match.

2. Deployment Considerations

o Can be deployed on cloud-based servers for large-scale applications or edge devices


for real-time processing.
o Optimization techniques like quantization and model pruning improve efficiency
for mobile and embedded applications.

This architecture ensures high accuracy in facial recognition while maintaining scalability
and real-time performance.

Methodology/Algorithm Details:
The dataset methodology defines the selection, preprocessing, and implementation of datasets
for training and evaluating the deep learning-based human face recognition model. This
includes dataset selection, preprocessing techniques, partitioning, augmentation, and
evaluation strategies.
1. Dataset Selection
The performance of a face recognition model depends on the quality and diversity of the
dataset. Commonly used datasets include:

• Labeled Faces in the Wild (LFW): A benchmark dataset with 13,000 images for face
verification.
• CASIA-WebFace: A large-scale dataset with over 10,000 identities for deep learning
training.

2. Dataset Preprocessing
Preprocessing ensures that images are standardized and ready for training. The key steps
include:

• Face Detection: Using algorithms like Haar Cascades, MTCNN, or Dlib’s HOG-
based detector to locate faces in images.
• Face Alignment: Aligning detected faces to ensure consistency in orientation.
• Cropping & Resizing: Extracting face regions and resizing them (e.g., 224×224
pixels) for CNN input.
• Normalization: Scaling pixel values between 0 and 1 or normalizing based on dataset
statistics to enhance model convergence.

3. Dataset Partitioning
To train and evaluate the model effectively, the dataset is divided into:

• Training Set (70-80%): Used for model learning.


• Validation Set (10-15%): Fine-tunes hyperparameters and prevents overfitting.
• Test Set (10-15%): Evaluates final model performance on unseen data.

This partitioning ensures that the model generalizes well to new faces.
4. Data Augmentation
Since deep learning models require large datasets, augmentation helps increase variability.
Common augmentation techniques include:

• Rotation & Flipping: Simulates different head poses.


• Brightness & Contrast Adjustment: Improves robustness to lighting variations.
• Gaussian Noise Addition: Prevents overfitting and improves generalization.

5. Evaluation Strategy
• Face Verification
(1:1 Matching): Determines if two images belong to the same •
Performance Metrics:
o Accuracy: Measures correct predictions. o False Acceptance Rate (FAR) &
False Rejection Rate (FRR): Measures system reliability.
o Receiver Operating Characteristic (ROC) Curve: Evaluates classification
performance.

Results:

Conclusion:
The implementation of deep learning-based human face recognition demonstrates the
effectiveness of Convolutional Neural Networks (CNNs) in accurately identifying and
verifying faces. By leveraging large-scale datasets and preprocessing techniques such as face
detection, alignment, and normalization, the model achieves improved generalization across
diverse conditions.

Through dataset partitioning and augmentation, the system enhances robustness against
variations in lighting, pose, and occlusions. Performance evaluation using accuracy, false
acceptance rate (FAR), and ROC curves ensures reliability in real-world applications.
The architecture supports both face verification (1:1 matching) and face identification (1:N
matching), making it suitable for security, surveillance, and authentication systems. While
deep learning improves recognition accuracy, challenges like privacy concerns and
adversarial attacks remain critical considerations.
Future advancements may include transformer-based models and edge computing for realtime
face recognition. Overall, deep learning has revolutionized face recognition, making it a vital
technology in biometric authentication and security systems.

You might also like