This repository provides the implementation of BREMOLA (Blind/Referenceless Model via Moving Spectrum and Laplacian Filter), a No-Reference Image Quality Assessment (NR-IQA) method tailored for autonomous driving environments. BREMOLA is designed to accurately assess image quality degradation, particularly blur, in real-time driving scenarios by leveraging a Fourier transform-based shifted spectrum and Laplacian filter for edge detection.
- Title: No-Reference Image Quality Assessment with Moving Spectrum and Laplacian Filter for Autonomous Driving Environment
- Authors: Woongchan Nam, Taehyun Youn, Chunghun Ha
- Published in: Vehicles
- DOI: 10.3390/vehicles7010008
- No-Reference IQA: Evaluates image quality without needing a reference image.
- Real-Time Adaptability: Designed for rapidly changing autonomous driving environments.
- Fourier Transform-based Shifted Spectrum: Quantifies image sharpness loss due to blur.
- Laplacian Filter Compensation: Reduces metric variance caused by environmental complexity.
- Robust to Driving Conditions: Tested on images from diverse locations (Dubai, Los Angeles, San Francisco, Seoul).
Before applying BREMOLA, NR-IQA methods based on moving spectrum analysis showed high variability in quality metric values due to environmental factors (lighting, number of objects, etc.), especially for low blur levels (1×1 to 5×5 filters). In particular, overlapping metric values between 1×1 (original) and 3×3 (mild blur) made it difficult to distinguish normal and degraded conditions. However, after applying BREMOLA, the Laplacian filter was used to compensate for image complexity, reducing variance in quality metrics and providing a more consistent decrease in values as blur increased, making it easier to identify degraded images.
PSNR and SSIM, as Full-Reference IQA (FR-IQA) methods, require an original reference image, making them unsuitable for autonomous driving. BRISQUE, a No-Reference IQA (NR-IQA) method, was highly sensitive to environmental changes, leading to inconsistent quality assessments influenced more by image complexity than actual blur. In contrast, BREMOLA is robust to environmental variations and provides a reliable measure of blur, making it better suited for real-time camera sensor monitoring in autonomous driving environments. 🚗💡
BREMOLA demonstrates higher reliability and accuracy in detecting blur degradation compared to traditional IQA metrics such as:
- PSNR (Peak Signal-to-Noise Ratio)
- SSIM (Structural Similarity Index)
- GMSD (Gradient Magnitude Similarity Deviation)
- BRISQUE (Blind/Referenceless Image Spatial Quality Evaluator)
| Method | Performance on Driving Images |
|---|---|
| BREMOLA | ✅ Stable, accurate, and robust |
| PSNR | ❌ Sensitive to noise, low robustness |
| SSIM | ❌ Requires reference images |
| GMSD | ❌ Unstable for varying conditions |
| BRISQUE | ❌ High variance in real-world driving |
The dataset consists of real-world driving images from various sources, covering:
- Different times of day
- Urban and highway scenarios
- Diverse environmental factors (buildings, vehicles, pedestrians, lighting conditions)
For privacy reasons, we do not provide the dataset here, but you can use publicly available driving footage from platforms like YouTube.
- Autonomous Vehicle Safety Monitoring
- Real-Time Camera Health Assessment
- Image Processing for ADAS (Advanced Driver Assistance Systems)
- Surveillance and Traffic Monitoring Systems
- Extending BREMOLA to handle other distortions (e.g., noise, motion blur).
- Improving real-time performance for embedded systems.
- Exploring deep-learning-based approaches for further enhancement.
If you find this work useful, please cite:
@article{nam2025bremola,
author = {Nam, Woongchan and Youn, Taehyun and Ha, Chunghun},
title = {No-Reference Image Quality Assessment with Moving Spectrum and Laplacian Filter for Autonomous Driving Environment},
journal = {Vehicles},
volume = {7},
year = {2025},
doi = {10.3390/vehicles7010008}
}