Anisotropic weighted edge-preserving pyramid for high dynamic range multi-exposure image fusion

Abstract

High dynamic range (HDR) imaging is critical for accurately representing scenes with large luminance variations, a challenge that is particularly pronounced in industrial environments containing objects with locally high reflectance. Conventional imaging techniques often fail to preserve essential details under such conditions, resulting in information loss in saturated regions. Multi-exposure image fusion (MEF) provides an effective approach; however, existing methods frequently exhibit detail degradation and halo artifacts. To address these limitations, a multi-scale image fusion framework is proposed, which leverages adaptive exposure, color gradients, and contrast. An adaptive exposure weight with fine-tuning factors, a Gaussian difference color gradient extractor, and image contrast are employed to derive feature weight maps. Halo artifacts are mitigated through the improvement of an anisotropic weighted guided filter, forming an anisotropic weighted edge-preserving pyramid (ASWEP) that refines the weight maps, followed by pyramid fusion to generate the final fused image. A dedicated industrial HDR image dataset is constructed to capture real-world industrial imaging conditions involving highly reflective materials. Extensive experiments on both the self-constructed industrial dataset and public dual- and triple-exposure datasets demonstrate superior detail preservation and effective halo suppression. Quantitative evaluations indicate consistent improvements over state-of-the-art methods across multiple metrics, including PSNR, MEF-SSIM, NMI, Qcv, SD, and VIF.