Related Experiment Video
Updated: Jun 26, 2026
07:13
Multimodal Cross-Device and Marker-Free Co-Registration of Preclinical Imaging Modalities
Published on: October 27, 2023
MDFAT: Interactive mask decoupling and frequency-adaptive transformer for multi-focus image fusion
Zhendong Xu1, Hao Zhai1, Zhi Zeng1
1School of Computer and Information Science, Chongqing Normal University, Chongqing, 401331, China.
Summary
This study introduces a novel network for multi-focus image fusion, improving global context modeling and feature separation. The frequency-adaptive transformer enhances efficiency and accuracy in creating a single, clear image.
Area of Science:
- Computer Vision
- Image Processing
Background:
- Multi-focus image fusion combines images with varied focal planes into one sharp image.
- Current methods struggle with global context and precise feature separation.
Purpose of the Study:
- To develop an advanced network for multi-focus image fusion.
- To overcome limitations in existing fusion techniques regarding context and feature handling.
Main Methods:
- Proposed a network integrating a fusion interactive mask decoupling mechanism and a frequency-adaptive transformer dual mechanism.
- Utilized Fourier transform for efficient frequency-domain attention computation and adaptive blocking.
- Implemented a dynamic mask-guided feature decoupling strategy for targeted foreground, background, and global context fusion.
Main Results:
- The proposed method demonstrated superior performance across objective metrics and subjective visual quality.
- Outperformed twelve state-of-the-art algorithms in systematic experiments on four datasets.
Conclusions:
- The novel network effectively models global context and accurately separates complementary features for superior multi-focus image fusion.
- The frequency-adaptive transformer and mask decoupling significantly enhance fusion performance and computational efficiency.