| 摘 要: 为了探讨图神经网络在处理异源图像配准任务中的应用和性能,为后续的图像融合或拼接等任务提供支持,通过综述现有文献,介绍了多种图神经网络模型及其在图像配准领域的应用,并对各种图神经网络架构进行了比较,评估了不同模型的性能。研究发现,图神经网络模型凭借其对图结构信息的有效利用能力及对节点属性信息的精细捕捉,在处理异源图像配准时展现出较传统方法更优的性能。通过对图像配准方法的系统研究,为解决异源图像配准任务面临的配准难度大和精度低的问题提供了新的技术思路。 |
| 关键词: 异源图像;图像配准;特征匹配;图神经网络;注意力机制 |
|
中图分类号: TP751
文献标识码: A
|
| 基金项目: 国家自然科学基金(32460440);甘肃省科技计划项目(20JR5RA032);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资助项目(GAU-QDFC-2020-08) |
|
| Overview of Heterogeneous Image Registration Methods Basedon Graph Neural Networks |
|
HUANG Dongfu, LIU Liqun
|
College of Inf ormation Science and Technology, Gansu Agricultural University, Lanzhou 730070, China
1073323120830@st.gsau.edu.cn; llqhjy@126.com
|
| Abstract: This research aims to explore the application and performance of Graph Neural Network (GNN) in addressing heterogeneous image registration tasks,providing support for subsequent tasks such as image fusion or stitching. By reviewing existing literature, various GNN models and their applications in the field of image registration are introduced, and a comparison of different GNN architectures is conducted to evaluate the performance of each model. The research reveals that GNN models demonstrate superior performance in handling heterogeneous image registration tasks compared to traditional methods, leveraging their ability to effectively utilize graph structural information and finely capture node attribute information. This systematic study of image registration methods offers new technical insights for addressing the challenges of low accuracy and high difficulty in heterogeneous image registration tasks. |
| Keywords: heterogeneous images; image registration; feature matching; GNN; attention mechanism |
