Scholarlink

<p>This paper proposes a novel generative adversarial network model named GF-GAN for fusing infrared and visible images. Traditional fusion methods often struggle with effectively merging complementary information from these two modalities. To address this, GF-GAN employs a dual Markov discriminator and computes gradient information as dynamic weights to guide the fusion process. Additionally, a gradient attention module is introduced to redistribute feature weights based on the relationship between current and surrounding gradients. Furthermore, dynamic information entropy is utilized to balance pixel losses, enabling the generator to adaptively produce fused images with rich gradient information. Experimental results on three publicly available datasets demonstrate that GF-GAN outperforms mainstream fusion methods in terms of both visual perception and objective evaluation metrics. The proposed approach holds promise for advancing high-level vision applications requiring comprehensive image information. The source code for the GF-GAN model is available at <a href="https://github.com/catjsjsj/GF-GAN">https://github.com/catjsjsj/GF-GAN</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Guided fusion of infrared and visible images using gradient-based attentive generative adversarial networks

  • 期刊论文

  • 发表日期：

    2025年1月23日

查看全文
  • Xingyue Zou,
  • Jiqiang Tang

摘要

This paper proposes a novel generative adversarial network model named GF-GAN for fusing infrared and visible images. Traditional fusion methods often struggle with effectively merging complementary information from these two modalities. To address this, GF-GAN employs a dual Markov discriminator and computes gradient information as dynamic weights to guide the fusion process. Additionally, a gradient attention module is introduced to redistribute feature weights based on the relationship between current and surrounding gradients. Furthermore, dynamic information entropy is utilized to balance pixel losses, enabling the generator to adaptively produce fused images with rich gradient information. Experimental results on three publicly available datasets demonstrate that GF-GAN outperforms mainstream fusion methods in terms of both visual perception and objective evaluation metrics. The proposed approach holds promise for advancing high-level vision applications requiring comprehensive image information. The source code for the GF-GAN model is available at https://github.com/catjsjsj/GF-GAN.