Infrared and Visible Image Fusion Based on Gaussian Weighted Standard Deviation Filter
TL;DR
The proposed GWSDF, theoretically underpinned by scale-space theory, is employed to achieve effective multi-scale image decomposition and achieves competitive performance against state-of-the-art fusion approaches in terms of detail retention, target saliency, texture integrity, and visual fidelity.
Abstract
Infrared and visible image fusion (IVIF) aims to generate a comprehensive and informative fused image by combining complementary thermal radiation and texture details from dual-modal source images. However, current fusion techniques still suffer from inadequate preservation of thermal targets and fine-grained details, as well as high computational complexity. To alleviate these problems, this paper proposes a novel IVIF algorithm based on the Gaussian weighted standard deviation filter (GWSDF). The proposed GWSDF, theoretically underpinned by scale-space theory, is employed to achieve effective multi-scale image decomposition. It reliably separates source images into base, salient, and detail layers, enabling effective separation of prominent thermal radiation targets from intricate visible texture details. Furthermore, the Fejér–Korovkin wavelet transform (FKWT) is adopted for base-layer fusion to maintain favorable spectral and structural consistency, contributing to visually plausible and clear fusion outputs. We conduct extensive experiments on three commonly used benchmark datasets: TNO, LLVIP, and M3FD. Both qualitative visual comparisons and quantitative objective evaluations demonstrate that the proposed method achieves competitive performance against state-of-the-art fusion approaches in terms of detail retention, target saliency, texture integrity, and visual fidelity. Moreover, the proposed method exhibits low computational complexity and fast execution speed among advanced non-learning fusion algorithms, making it highly practical for real-time fusion-oriented applications.