| 摘 要: 传统的红外和可见光图像融合方法在不同环境中的应用效果表现不稳定。针对此问题,提出了一种基于自注意力机制和RSA(Rivest-Shamir-Adleman)加密的图像融合算法。首先,在图像融合部分提出了一种自适应权重学习模块,该模块可以实现动态权重分配。其次,提出了一种动态密钥管理机制并将其融入RSA加密算法中,将融合后的图像进行加密,提高图片的安全性。实验结果表明,与Densefusion、FusionGAN、IFCNN、TarDAL四种融合方法相比,所提方法的客观评价指标MI、VIF、SSIM、FMIdct、Qabf分别平均提升了16.35%、26.56%、14.58%、18.27%、20.79%。此外,对加密后的图像进行安全性分析,实验表明该算法具有较高的安全性。 |
| 关键词: 自注意力机制;RSA加密算法;图像融合算法;自适应权重学习;图像加密 |
|
中图分类号: TP751;TP309.
文献标识码: A
|
| 基金项目: 国家自然科学基金(32460440);甘肃省高校教师创新基金项目(2023A-051);甘肃农业大学青年导师基金项目(GAU-QDFC-2020-08);甘肃省科技计划项(20JR5RA032) |
|
| Image Fusion Algorithm Based on Self-Attention Mechanism And RSA Encryption |
|
WU Zongxiang, LIU Liqun
|
(College of Inf ormation Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
wuzx@st.gsau.edu.cn; llqhjy@126.com
|
| Abstract: Traditional methods for fusing infrared and visible light images demonstrate unstable performance across different environments. To address this issue, this paper proposes an image fusion algorithm based on selfattention mechanism and RSA (Rivest-Shamir-Adleman) encryption. Firstly, an adaptive weight learning module is introduced in the image fusion process that enables dynamic weight allocation. Secondly, a dynamic key management mechanism is proposed and it is integrated into the RSA encryption algorithm to encrypt the fused image, enhancing its security. Experimental results show that, compared to four existing fusion methods—Densefusion, FusionGAN, IFCNN,and TarDAL—the proposed method achieves average improvements of 16.35% , 26.56% , 14.58% , 18.27% , and 20.79% in objective evaluation metrics MI, VIF, SSIM, FMIdct, and Qabf, respectively. Moreover, a security analysis of the encrypted images demonstrates that the algorithm possesses a high level of security. |
| Keywords: self-attention mechanism; RSA encryption algorithm; image fusion algorithm; adaptive weight learning; image encryption |
