Fine-grained multi-focus image fusion based on edge features.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Multi-focus image fusion aims to combine images with varying focal planes into a single all-in-focus image.
• Existing fusion methods often struggle with retaining fine details from the source images.
• This limitation impacts the practical utility of fused images in various applications.

Purpose of the Study:

• To develop an advanced multi-focus image fusion architecture.
• To improve the retention of detail information from original images in the fusion process.
• To achieve superior fusion performance compared to existing methods.

Main Methods:

• A two-stage fusion architecture was designed, incorporating an encoder-decoder network.
• The encoder-decoder utilized a polarized self-attention module and DenseNet structure for enhanced image reconstruction.
• A fusion strategy based on edge feature maps was employed in the second stage to focus on detail preservation.

Main Results:

• The proposed algorithm demonstrated advanced performance in subjective and objective evaluations against nine classical fusion algorithms.
• The fused images exhibited significantly better information retention from the original source images.
• The edge-based fusion strategy effectively enhanced attention to critical detail information.

Conclusions:

• The novel two-stage fusion architecture effectively addresses the detail information retention issue in multi-focus image fusion.
• The integration of polarized self-attention and edge feature maps leads to superior fusion quality.
• This method offers a promising solution for applications requiring high-fidelity fused images.