An α-Matte Boundary Defocus Model-Based Cascaded Network for Multi-focus Image Fusion

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Limited depth of field in single cameras necessitates multi-focus image fusion.
• Existing fusion methods struggle with clarity at focused/defocused boundaries (FDB).

Purpose of the Study:

• To propose a novel α-matte boundary defocus model for realistic training data generation.
• To develop a cascaded network (MMF-Net) for improved multi-focus image fusion near FDBs.

Main Methods:

• Development of an α-matte boundary defocus model to simulate defocus spread effects.
• Design and training of a cascaded convolutional network (MMF-Net) with initial and boundary fusion subnets.
• Generation of training data using the proposed α-matte model.

Main Results:

• The MMF-Net effectively refines fusion results specifically around the FDB.
• The proposed α-matte model accurately captures defocus spread, enhancing training data realism.
• MMF-Net demonstrates superior qualitative and quantitative performance compared to state-of-the-art methods.

Conclusions:

• The novel α-matte boundary defocus model enables precise simulation of defocus effects.
• MMF-Net achieves state-of-the-art multi-focus image fusion, particularly improving clarity at focused/defocused boundaries.