A Variational Multi-Scale Model for Multi-Exposure Image Fusion.

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Multi-exposure image fusion (MEF) is crucial for generating High Dynamic Range (HDR) images.
• Existing methods often struggle with detail preservation and artifact reduction.

Purpose of the Study:

• To propose a novel variational model for MEF.
• To enhance HDR image quality through detail-based decomposition and reconstruction.
• To address limitations in current MEF techniques.

Main Methods:

• Developed a unified variational framework integrating decomposition and reconstruction.
• Employed Tikhonov regularization for base layer modeling.
• Utilized an efficient design for detail layer extraction and multi-scale techniques for artifact removal.
• Applied the Alternating Direction Method of Multipliers (ADMM) for numerical optimization.

Main Results:

• The proposed model demonstrates superior performance in visual quality and objective metrics.
• Achieved 1%-10% improvement in Natural Image Quality Evaluator (NIQE) for real image fusion.
• Showcased 13%-20% improvement in Peak Signal-to-Noise Ratio (PSNR) for synthetic image fusion.

Conclusions:

• The novel variational model offers significant improvements in MEF.
• The integrated approach effectively preserves details and reduces artifacts.
• The method provides a robust solution for generating high-quality HDR images.