EAST WEST INSTITUTE OF TECHNOLOGY

Affiliated to VTU and Approved by AICTE, New Delhi

Department of Computer Science and Engineering

Technical Seminar

“FACE RECOGNITION USING CNN"

By, Under the Guidance of

Usha Rani P S Prof.Nalini B M
1EW20CS143 Assistant Professor,
CSE, EWIT Department of CSE, EWIT
CONTENTS

 INTRODUCTION
 TECHNOLOGY USED
 ALGORITHMS USED
 IMPLEMENTATION
 RESULTS
 ADVANTAGES AND DISADVANTAGES
 APPLICATIONS
 CONCLUSION
 REFERENCES
INTRODUCTION
 In our increasingly digital world, the ability to accurately identify individuals is becoming
crucial for various applications. Face recognition technology offers a powerful solution,
and Convolutional Neural Networks (CNNs) are at the forefront of this advancement.
These networks excel at extracting intricate features from facial images, enabling them to
recognize individuals with remarkable accuracy. Today, we'll explore the fascinating
world of CNN-based face recognition, delving into its inner workings and
potential applications.
Architecture of CNN
TECHNOLOGY USED
 Convolutional Layers: These layers apply filters to extract features like edges,
textures, and shapes, capturing the essence of a face.

 Pooling Layers: They reduce data size and enhance feature representation by
summarizing the information extracted by convolutional layers.

 Activation Functions: These introduce non-linearity, allowing the network to learn

complex relationships between features. Popular examples include ReLU and
Leaky ReLU.

 Fully Connected Layers: These layers combine the extracted features from previous
layers and make the final classification decision, identifying the
individual in the image.
 Algorithm
1. Data Preprocessing:
Input: A dataset of facial images with labels indicating the identity of each individual.
Steps:
Data Collection: Gather images from various sources, ensuring diversity in age, ethnicity, pose, and lighting
conditions.
Preprocessing: Resize and crop images to a standard size. Normalize pixel values for consistency . Apply
data augmentation techniques like flipping, rotation, and color jittering to increase dataset size and diversity.
2. Model Definition :
Input: Preprocessed facial images.
Output: Predicted identity label for each image.
Architecture:
Convolutional Layers : Use multiple convolutional layers with different filter sizes and activation functions (e.g.,
ReLU) to extract features like edges, textures, and shapes from the images.
 Pooling Layers : Reduce data size and enhance feature representation by summarizing information from convolutional
layers
Fully Connected Layers : Combine extracted features and learn the mapping between features and class labels. Use
multiple fully connected layers with dropout layers (optional) to prevent overfitting.
Output Layer : Use a softmax activation function to predict the probability of each class (individual) for the input image.
3. Training :
Input: Preprocessed images, corresponding labels, and hyperparameters (learning rate, optimizer, etc.).
Steps:
Split the data: Divide the dataset into training, validation, and testing sets.
Initialize the model: Set random weights and biases for the network layers.
4. Testing:
Input: Preprocessed images from the testing set.
Output: Predicted identity labels for the test images.
Steps: Feed the preprocessed test images through the trained model. Obtain the predicted labels for each image based
on the highest probability output from the softmax layer. Evaluate the model's performance using metrics like accuracy,
precision, recall, and F1-score.
 IMPLEMENTATION
1. Data Preprocessing:
Data Collection: Gather a dataset of facial images with labels indicating the identity of each individual. Public
datasets like MNIST, Labeled Faces in the Wild (LFW), or CelebA are commonly used.
Preprocessing: Clean the data by addressing missing values, outliers, and inconsistencies. Apply techniques like
normalization, resizing, and cropping to ensure consistency across images.
Data Augmentation (Optional): Artificially increase the size and diversity of your dataset by applying techniques like
random cropping, flipping, rotation, and color jittering. This helps the model generalize better to unseen variations.
2. Model Architecture :
Choose a CNN architecture: Popular choices include VGGFace, ResNet, or Inception, which are pre-trained on large
image datasets and can be fine-tuned for face recognition.Define the network layers: The architecture typically consists of
convolutional layers for feature extraction, pooling layers for dimensionality reduction, activation functions for non-
linearity, and fully connected layers for classification.Loss function and optimizer: Select appropriate loss function (e.g.,
cross-entropy) and optimizer (e.g., Adam) to guide the training process and update model parameters.
3. Training :
Split the data: Divide the preprocessed dataset into training, validation, and testing sets. The training set is used to
train the model, the validation set is used to monitor performance during training and prevent overfitting, and the testing
set is used for final evaluation after training is complete.
Train the model: Feed the training data into the model and iteratively update its weights based on the chosen
optimizer and loss function. Monitor the validation accuracy and loss to ensure the model is learning effectively and avoid
overfitting.
Hyperparameter tuning: Experiment with different hyperparameters like learning rate, batch size, and number of
epochs to optimize the model's performance.
4. Testing:
Feed the preprocessed test images through the trained model. Obtain the predicted labels for each image based on
the highest probability output from the softmax layer. Evaluate the model's performance using metrics like accuracy,
precision, recall, and F1-score
RESULTS
Evaluation Metrics :
Accuracy: The percentage of correctly classified faces.
Precision: The proportion of true positives among predicted positives.
Recall: The proportion of true positives identified by the model.
F1-score: A harmonic mean of precision and recall, combining both metrics.
Visualization:
Confusion matrix: Visualize the distribution of correct and incorrect predictions for each. individual.
ROC curve: Analyze the trade-off between true positive rate and false positive rate.
Interpretation: Analyze the results to understand the model's strengths and weaknesses. Identify potential biases or limitations in the dataset or model architecture. Consider techniques like saliency maps to
visualize which parts of the image contribute most to the model's decision.
GRAPHICAL RESULT ANALYTICS

fig: evolution matrix fig: confusion matrix

fig: ROC Curve

 ADVANTAGES
 High Accuracy

 Robustness to variations

 Learns from raw data

 Data Efficiency
 DISADVANTAGES
► Computational cost
► Data requirements
► Privacy concerns
► Bias
 APPLICATIONS
 Security and surveillance
 Finance and retail
 Healthcare
 Entertainment and social media
 Education
 Mobile devices
 CONCLUSION
 In conclusion, face recognition using Convolutional Neural Networks (CNNs) has proven to be highly
effective in accurately identifying and verifying faces in various applications. The ability of CNNs to
automatically learn hierarchical features from facial images contributes to robust and reliable recognition
performance. However, challenges such as variations in lighting conditions, pose, and occlusions persist,
prompting ongoing research to enhance the robustness of CNN-based face recognition systems. Despite
these challenges, CNNs remain a cornerstone in the advancement of facial recognition technology,
offering promising results for improved accuracy and real-world applicability.
 REFERENCE

[1] C. Low, A. B. Teoh and C. Ng. “Multi-Fold Gabor, PCA, and ICA Filter Convolution Descriptor for Face
Recognition" [J]. IEEE Transactions on Circuits and Systems for Video Technology, 2019, 29(1):115-129.
[2 ] Mei Wang, Weihong Deng. Deep face recognition: A survey , Neurocomputing[J].2021(429): 215-244.
[3] Ou Yang, W. etal. Deep ID-Net: Deformable Deep Convolutional Neural Networks for Object Detection. [J].IEEE
transactions on pattern analysis and machine intelligence.2017, 39(7):1320-1334.
[4] Liu W , Wen Y , Yu Z , et al. Sphere Face : Deep Hypersphere Embedding for Face Recognition[C]// 2017 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017
[5] H. Wang et al. Cos Face: Large Margin Cosine Loss for Deep Face Recognition[C]// 2018 IEEE/CVF
Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, 2018, pp. 5265-5274.
THANK YOU