A PDE-Driven Laplacian Multi-Exposure Fusion for Structure-Preserving Image

Abstract

Multi-Exposure Image Fusion (MEF) aims to fuse a single visually symmetrical image, which combines the luminosity and preserves the structural data of a scene at different exposure levels. Traditionally, Laplacian Pyramid (LP) fusion methods rely on Gaussian smoothing, which diffuses intensity isotropically, resulting in blurred edges, halo artifacts, and loss of fine detail. To overcome these limitations, the paper proposes a method comprising Mean Curvature and Laplacian Pyramid that combines curvature-based edge preservation with multiscale structural decomposition. Mean Curvature Filtering (MCF), formulated on curvature flow principles, acts as an anisotropic smoothing operator that preserves edges and geometric discontinuities while suppressing low-frequency noise. The MCF is integrated into the pyramid construction to ensure structure-aware decomposition, producing curvature-sensitive base and detail layers. A covariance-based adaptive weighting strategy is employed for detail layer fusion, enhancing locally salient textures, while the base layer is averaged to maintain consistent illumination. The fused image is reconstructed through recursive MCF-based expansion to achieve seamless integration of global brightness and multiscale structural details. Experimental evaluation indicates that the proposed framework achieves enhanced sharpness, natural contrast, and reduced artifacts by effectively balancing noise suppression with edge fidelity. The approach thus provides a mathematically consistent and perceptually coherent formulation for multi-exposure fusion using curvature-driven multiscale representation.