Robust Multi-Exposure Image Fusion: A Structural Patch Decomposition Approach.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Multi-exposure image fusion (MEF) aims to combine multiple images of the same scene taken at different exposures into a single, high-quality image.
• Existing MEF methods often struggle with ghosting artifacts, especially in dynamic scenes with motion.
• Pixel-wise methods typically require post-processing to mitigate visual artifacts.

Purpose of the Study:

• To develop a robust and efficient multi-exposure image fusion method that minimizes ghosting effects.
• To introduce a patch decomposition technique that inherently handles motion inconsistencies.
• To improve the visual quality and color vividness of fused images.

Main Methods:

• A structural patch decomposition approach is proposed, separating image patches into signal strength, signal structure, and mean intensity components.
• These components are fused independently, and then reconstructed into a final fused image.
• The signal structure component's directional information is utilized for ghost removal without explicit motion estimation.

Main Results:

• The proposed method outperforms 12 existing MEF algorithms on static scenes.
• It consistently produces high-quality fused images with minimal ghosting artifacts on dynamic scenes.
• The algorithm demonstrates a lower computational cost compared to state-of-the-art deghosting schemes.

Conclusions:

• The structural patch decomposition approach offers an effective and efficient solution for multi-exposure image fusion.
• The method successfully addresses ghosting artifacts in dynamic scenes, enhancing image quality and color.
• It provides a computationally advantageous alternative to existing MEF and deghosting techniques.