Efficient Multiscale Gradient-Domain Filtering for Image and Video Dehazing with Enhanced Temporal Coherence

Area of Science:

• Computer Vision
• Image Processing

Background:

• Atmospheric scattering degrades image clarity, posing challenges for computer vision tasks.
• Existing dehazing methods often exhibit high computational costs and can lose important gradient details.
• Video dehazing specifically suffers from flickering artifacts, limiting real-world applicability.

Purpose of the Study:

• To propose an efficient and high-quality dehazing technique for both images and videos.
• To address limitations of existing methods, including computational cost, detail loss, and flickering artifacts.
• To enhance texture and edge retention while maintaining temporal coherence in video dehazing.

Main Methods:

• A multiscale gradient-domain weighted guided image filter (GWGIF) is proposed for refined transmission map estimation.
• Minimum Preserving Subsampling (MPS) is utilized for efficient atmospheric parameter estimation and complexity reduction.
• A Gradient-Based Correlation Factor (GCF) is introduced to mitigate flickering artifacts in video dehazing.

Main Results:

• The proposed method achieved superior perceptual quality with PIQE, NIQE, and BRISQE scores of 26.98, 2.78, and 20.18, respectively.
• High temporal coherence was demonstrated in video dehazing with a Mean Square Error (MSE) deviation of 0.003.
• Significant reduction in flickering artifacts was observed in video dehazing compared to existing approaches.

Conclusions:

• The proposed GWGIF-based dehazing technique offers enhanced efficiency and visual quality for both image and video processing.
• The method effectively preserves gradient details and improves texture/edge retention.
• Its suitability for real-time applications like autonomous driving and surveillance is highlighted due to its performance and temporal stability.