Researching | Infrared‒Visible Image Registration and Fusion Based on Dynamic Convolutional Dual Attention Network

Acta Optica Sinica (Online), Volume. 2, Issue 17, 1711002(2025)

Infrared‒Visible Image Registration and Fusion Based on Dynamic Convolutional Dual Attention Network

Keyang Xia¹, Shui Yu², Qian Ni², Qi Li², Hongna Zhu^1,2、*, and Bangji Wang^1,2、**

Author Affiliations

¹School of Information Science and Technology, Southwest Jiaotong University, Chengdu 610031, Sichuan , China

²School of Physical Science and Technology, Southwest Jiaotong University, Chengdu 610031, Sichuan , China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (12)

References (32)

Paper Information

Figures & Tables(12)

Fig. 1. Schematic diagram of affine transformation

Download full size

View in Article

Fig. 2. Structure of RRFNet-DCDAN

Download full size

View in Article

Fig. 3. Structure of registration module

Download full size

View in Article

Fig. 4. Structure of fusion module

Download full size

View in Article

Fig. 5. Structure of dual attention mechanism

Download full size

View in Article

Fig. 6. Structure of dynamic convolution^[25]

Download full size

View in Article

Fig. 7. Results of the registration network. (a) Source image; (b) registered image

Download full size

View in Article

Fig. 8. Comparison of experimental results on MSRS dataset

Download full size

View in Article

Fig. 9. Comparison of experimental results on RoadScene dataset

Download full size

View in Article

Table 1. Comparison of results from MSRS dataset

View in Article

Table 1. Comparison of results from MSRS dataset

Method	MI	VIF	SD	MS-SSIM	$Q^{A B / F}$
GTF	2.5713	0.5161	6.2107	0.2054	0.3428
DenseFuse	3.1015	0.7654	7.9596	0.4363	0.4905
U2Fusion	3.8152	0.6794	7.0493	0.5972	0.5361
IFCNN	3.0176	0.7161	7.1429	0.6638	0.5718
RRFNet	3.9859	1.0763	8.0985	0.6908	0.5947
RRFNet+DCA	3.8147	1.0926	8.0976	0.7063	0.5961
RRFNet+DAN	4.3195	1.1806	8.2350	0.7384	0.6027
RRFNet-DCDAN	4.1674	1.2153	8.2297	0.7509	0.6085

Table 2. Comparison of results from RoadScene dataset

View in Article

Table 2. Comparison of results from RoadScene dataset

Method	MI	VIF	SD	MS-SSIM	$Q^{A B / F}$
GTF	2.1148	0.5726	5.8295	0.1893	0.2973
DenseFuse	2.0512	0.7248	6.7918	0.4142	0.4467
U2Fusion	3.6597	0.7315	6.0294	0.5083	0.4976
IFCNN	3.4976	0.7108	6.2853	0.5597	0.5243
RRFNet	3.7934	1.0721	7.2307	0.5851	0.5872
RRFNet+DCA	3.8024	1.0950	7.3284	0.5973	0.5922
RRFNet+DAN	4.1051	1.0831	7.4863	0.6046	0.6094
RRFNet-DCDAN	4.0017	1.1608	7.4952	0.6098	0.6030

Table 3. Comparison of processing time for images with different resolutions
View table
View in Article
Table 3. Comparison of processing time for images with different resolutions
Algorithm 320 pixel×240 pixel 640 pixel×480 pixel 1280 pixel×960 pixel
RRFNet 0.75 1.07 1.73
RRFNet+DCA 0.59 0.86 1.37

Tools

Get Citation

Copy Citation Text

Keyang Xia, Shui Yu, Qian Ni, Qi Li, Hongna Zhu, Bangji Wang. Infrared‒Visible Image Registration and Fusion Based on Dynamic Convolutional Dual Attention Network[J]. Acta Optica Sinica (Online), 2025, 2(17): 1711002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Computational Optics

Received: Jul. 23, 2025

Accepted: Jul. 31, 2025

Published Online: Sep. 5, 2025

The Author Email: Hongna Zhu (hnzhu@swjtu.edu.cn), Bangji Wang (bangjiw@163.com)

DOI:10.3788/AOSOL250504

CSTR:32394.14.AOSOL250504

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology