This repository contains the code and resources for the paper "Real-Time Weather Image Classification with SVM: A Feature-Based Approach". The project uses Support Vector Machine (SVM) to classify weather conditions in images into four categories: rainy, low light, haze, and clear.
- Brightness
- Saturation
- Noise Level
- Blur Metric
- Edge Strength
- Motion Blur
- Local Binary Patterns (LBP) mean and variance for radii 1, 2, and 3
- Edges mean and variance
- Color histogram mean and variance for blue, green, and red channels
The SVM model achieved an accuracy of 92.8%, surpassing typical benchmarks for classical machine learning methods.
add_artificial_rain.py: Script to add artificial rain to images.simulate_low_light.py: Script to simulate low light conditions in images.add_haze.py: Script to add haze to images.weather_image_classification_svm.py: Script to extract features from images and classify them using SVM.
The dataset used for this project can be downloaded from the following link:
python add_artificial_rain.py /path/to/source_folder /path/to/destination_folder --rain_intensity 1500
To simulate low light conditions on images, use the following command: python simulate_low_light.py /path/to/input_folder /path/to/output_folder
To add haze to images, use the following command: python add_haze.py /path/to/image.jpeg /path/to/output_folder --betas 0.05 0.06 0.07 --A 0.5
To classify weather images using the SVM model, use the following command: python weather_image_classification_svm.py /path/to/dataset