MMFuse: A multi‐scale infrared and visible images fusion algorithm based on morphological reconstruction and membership filtering. Issue 4 (24th November 2022)

• Record Type: Journal Article
• Authors: Zhao, Liangjun; Yang, Hao; Dong, Linlu; Zheng, Liping; Asiya, Manlike; Zheng, Fengling
• Is Part Of: IET image processing. Volume 17:Issue 4(2023)
• Page Start: 1126
• Page End: 1148
• Publication Date: 2022-11-24
• Subjects: Image fusion -- Multi‐scale transformation -- Fuzzy c‐means clustering(FCM) -- Morphological reconstruction (MR); Image processing -- Periodicals; 621.36705
• DOI: 10.1049/ipr2.12701
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract: This study proposes a multi‐scale transformation method based on morphological reconstruction and membership filtering, termed as MMFuse, to fuse infrared and visible images. This method employs a fuzzy c‐means clustering algorithm for multi‐scale decomposition by introducing morphological reconstruction operations and modifying member partitions to ensure noise resistance and image detail preservation. In addition, the MMFuse utilises the image attributes of layers as their fusion weights at each scale for adaptive feature fusion, which reduces the difficulty of manual adjustment of fusion weights. Moreover, on the basis of histogram enhancement, a visible image enhancement method is proposed, which can help exploit additional texture details in low‐light visible images and transfer these details to the fused image. The experiments performed on public datasets indicates that the MMFuse can generate sharp and clean fused images with high robustness and good fusion results for the images corrupted by different noises. Moreover, the results of this method appear as high‐quality visible images with clear highlighted infrared targets. Abstract : We proposed a multi‐scale IR/VIS fusion method based on morphology reconstruction and membership filtering. It can keep both the thermal radiation and the texture details in the source images. It is able to provide good fusion results for images corrupted by different types of noise. It uses the inherent properties of layers as fusion weights, reducing the difficulty of adjusting fusion weights manually. Our results look like high‐quality visible images with clear highlighted infrared targets. We also proposed a visible image enhancement method based on histogram enhancement.