Multifocus Image Fusion via LS-DTNP in the NSST Domain

Multifocus image fusion (MFIF) is a process that aims to integrate multiple partially focused images to generate a single, all-in-focus image, thereby effectively extending the depth of field (DoF) of optical lenses. However, existing fusion methods often neglect the spatial correlation of pixels, leading to artifacts at the boundaries between focused and defocused regions and a lack of spatial consistency in fusion decision maps. To address this challenge, this article proposes an MFIF framework based on the linking synaptic dynamic threshold neural P (LS-DTNP) system. First, the nonsubsampled shearlet transform (NSST) is employed to decompose the source images into feature maps at various scales and directions. Next, to overcome the artifacts and noise associated with conventional methods, the LS-DTNP system is utilized to fuse the detail information within the high-frequency subbands, while the low-frequency subbands are merged using a fusion rule based on a local smoothing operator and multiscale Laplacian energy (LSO-MSLE). Finally, the resulting fused image is reconstructed via the inverse NSST. Comprehensive experiments on three public datasets-Lytro, MFFW, and MFI-WHU-demonstrate that, compared with 18 state-of-the-art (SOTA) fusion algorithms, the proposed method exhibits favorable performance across eight objective metrics. The results confirm its ability to generate fused images with superior visual quality and richer details, effectively showcasing the application potential of the LS-DTNP system in the field of image fusion.