A ghost-free multi-exposure image fusion using adaptive alignment for static and dynamic images

Abstract

Multi-Exposure image Fusion (MEF) blends images with varying exposures to construct a well-exposed outcome that retains all essential details. While many MEF techniques are effective, the dynamic image sets, where movements are present, pose challenges during fusion, leading to severe artifacts. Existing approaches inherently rely on the median image to align image sets before fusion for rectifying this crisis. However, the uncertainty caused by limited datasets and distorted median image during alignment is an ongoing critical issue in the domain. The proposed method presents a novel MEF framework, introducing a newly developed adaptive alignment technique and a unique Singular Value Decomposition (SVD) weight map, specifically designed to handle dynamic image sets. This strategy efficiently aligns the input images using a qualified reference image and performs pyramidal fusion using SVD along with adaptive well-exposedness, and contrast weight maps. This effectively handles both dynamic and static images, outperforming existing MEF techniques in visual analysis and empirical tests. Furthermore, significant performances from the execution time, pixel intensity analysis, and infrared-visible image fusion analysis confirm the practicality of our approach. The proposed methodology reinforces MEF's vital role in image processing applications such as medical imaging, surveillance, and remote sensing.