Face Mask Detection Project

This project involves building a Convolutional Neural Network (CNN) to classify images as either showing a person wearing a mask or not wearing a mask. It was developed using Python and various machine learning and image processing libraries. Below is a breakdown of the project:

1. Dataset Preparation

• Data Source: The dataset is downloaded from Kaggle, containing images of people with and without masks.
• Data Organization: The dataset is split into two main folders: with_mask and without_mask, containing labeled images.

2. Data Preprocessing

• Loading and Resizing Images: Images are loaded using PIL and resized to 128x128 pixels.
• Labeling: Images from the with_mask folder are labeled 1, and images from the without_mask folder are labeled 0.
• Normalization: Image data is normalized by dividing pixel values by 255 to scale them between 0 and 1.

3. Splitting Data

• The dataset is split into training and testing sets using train_test_split from sklearn with an 80-20 ratio.

4. Model Architecture

• Layers:
  • Conv2D and MaxPooling2D for feature extraction.
  • Flatten for converting 2D data to 1D.
  • Dense layers with ReLU activation and Dropout for regularization.
  • Final Dense layer with sigmoid activation for binary classification.
• Compilation:
  • Optimizer: Adam
  • Loss function: sparse_categorical_crossentropy
  • Metrics: accuracy

5. Training the Model

• The model is trained for 5 epochs with an 80-20 validation split.
• Training metrics (accuracy and loss) are tracked and displayed.

6. Evaluation

• The model achieves an accuracy of ~90.5% on the test data.
• Plotting: Training and validation loss/accuracy are plotted to visualize model performance.

7. Image Prediction

• Users can input an image path for prediction.
• The input image is read using cv2, resized, normalized, and reshaped before being passed to the model for prediction.
• The model outputs whether the person in the image is wearing a mask or not.

8. Libraries Used

• numpy, matplotlib, PIL, cv2, tensorflow, sklearn

9. Sample Result

• Training accuracy: ~89.2%
• Validation accuracy: ~91.8%
• Test accuracy: ~90.5%

10. Visualization

• Training and validation loss/accuracy curves are plotted for better insights.

This project serves as a practical implementation for learning about CNNs in computer vision and image classification tasks.