MDbFusion++: A Visible and Infrared Image Fusion Framework Capable for Motion Deblurring

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Existing image fusion techniques struggle with motion blur in source images, leading to distorted details and degraded performance in subsequent high-level tasks.
• Motion blur, caused by object movement, is a significant challenge in acquiring clear visual and infrared imagery for fusion applications.

Purpose of the Study:

• To propose a novel framework, MDbFusion++, capable of simultaneously performing image fusion and motion deblurring.
• To develop a mutually reinforcing approach where deblurring enhances fusion and fused images provide feedback for improved deblurring.

Main Methods:

• MDbFusion++ utilizes a coarse-to-fine image restoration strategy with two key components: a coarse deblurring part (CDP) and a fine deblurring and fusion part (FDFP).
• CDP employs an adaptive weights module (AWM) to leverage infrared features for compensating motion-blurred visible images.
• FDFP uses a multi-domain enhancement module (MEM) to restore fine visible features and fuse images across spatial and frequency domains.

Main Results:

• The proposed network effectively performs both image deblurring and fusion.
• Deblurred visible features provide clearer information, significantly improving fusion outcomes.
• Fused images offer gradient feedback, further enhancing the deblurring performance.

Conclusions:

• MDbFusion++ demonstrates superior performance in both deblurring and fusion tasks compared to existing methods.
• The framework offers distinct advantages for applications requiring high-quality fused images from motion-blurred sources.