Outdoor images often suffer from reduced visibility due to haze, fog, or atmospheric scattering. This project proposes an Artificial Multiple Exposure Fusion (AMEF) method for single-image dehazing. Unlike traditional techniques that rely on depth estimation or inversion of haze models, AMEF underexposes the hazy image through a series of gamma corrections. The set of artificially exposed images is then fused using a multi-scale Laplacian blending scheme, producing a haze-free image.
Key outcomes:
- Removes haze effectively without depth maps.
- Provides both qualitative and quantitative improvements.
- Open-source implementation available for reproducibility.
- Image dehazing improves degraded visibility caused by haze or fog.
- Conventional methods rely on physical haze models, depth estimation, or image priors.
- Proposed Approach: Treat haze removal as a spatially-varying contrast and saturation enhancement problem.
- Input: A hazy image.
- Gamma Correction: Create a set of artificially underexposed versions.
- Fusion Strategy:
- Use Laplacian pyramid decomposition.
- Fuse best-quality regions from each exposure.
- Generate a single haze-free output.
Advantages:
- Avoids explicit depth or transmission estimation.
- Efficient and robust against varying haze density.
- Preserves color and contrast without artifacts.
- Metrics: Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM).
- Example values:
- PSNR = 22.02
- SNR = 17.56
- Subjective Evaluation: AMEF achieves results comparable to Dark Channel Prior (DCP) while avoiding color distortions.
- Quantitative Evaluation: AMEF ranks second-best in SSIM performance, outperforming several state-of-the-art techniques.
- Efficiency: Fastest runtime among compared dehazing methods.
- Autonomous driving and surveillance under poor visibility.
- Remote sensing and aerial imagery.
- Medical imaging and astronomy.
- Web mapping and land-use planning.
- Developed a robust, efficient, and high-quality dehazing technique (AMEF).
- Produces haze-free images without requiring depth estimation.
- Future scope: Apply AMEF principles to other image enhancement tasks such as illumination correction and contrast enhancement.