Robust Match Fusion Using Optimization.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Exposure fusion aims to combine multiple images with varying exposures into a single, well-exposed image.
• Dynamic scenes with moving objects pose challenges for traditional fusion methods, often resulting in ghosting artifacts.
• Existing methods frequently require image registration (alignment) before fusion, adding computational complexity.

Purpose of the Study:

• To develop a novel patch-based algorithm for exposure fusion that accurately handles moving objects in dynamic scenes.
• To eliminate the need for pre-alignment of multiple exposure image sequences.
• To achieve ghosting-free fusion results superior to current state-of-the-art techniques.

Main Methods:

• A uniform iterative approach is employed to match corresponding patches across different exposure images.
• A new patch-based energy function is optimized to enhance matching accuracy, considering exposure time variations.
• The random walker algorithm is utilized for efficient patch-based exposure fusion, preserving moving object details.

Main Results:

• The proposed algorithm successfully fuses multiple exposure image sequences of dynamic scenes without ghosting artifacts.
• Experimental results demonstrate visually pleasing fusion outcomes, outperforming existing exposure fusion and tone mapping algorithms.
• The method effectively preserves moving objects, a significant improvement over techniques prone to ghosting.

Conclusions:

• This patch-based approach represents a novel solution for exposure fusion in dynamic scenes, specifically addressing moving object preservation.
• The algorithm's ability to perform fusion without image registration simplifies the process and enhances efficiency.
• The developed method offers a robust and effective way to generate high-quality, artifact-free fused images from dynamic scene sequences.