DDcGAN: A Dual-discriminator Conditional Generative Adversarial Network for Multi-resolution Image Fusion

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Image fusion aims to combine information from multiple sources.
• Existing methods struggle with fusing images of different resolutions, leading to information loss.

Purpose of the Study:

• To propose a novel end-to-end model for fusing infrared and visible images of varying resolutions.
• To enhance the quality of fused images by preserving both thermal radiation and texture details.

Main Methods:

• Introduced a dual-discriminator conditional generative adversarial network (DDcGAN).
• DDcGAN employs a generator and two discriminators in an adversarial game.
• Incorporated a content loss and specific constraints for fusing different resolution images.

Main Results:

• The DDcGAN successfully fuses infrared and visible images, preserving thermal radiation and texture.
• The model effectively handles images of different resolutions, avoiding information blurring or loss.
• Demonstrated superior performance over state-of-the-art methods in qualitative and quantitative evaluations.

Conclusions:

• DDcGAN offers a robust solution for infrared and visible image fusion, especially with varying resolutions.
• The method shows potential for fusing other multi-modality medical images.
• The proposed approach significantly advances the field of image fusion technology.