|本期目录/Table of Contents|

[1]陈礼航,曾祥进*,熊瑞峰,等. 基于局部特征变换器的多尺度自适应特征图像匹配算法 [J].武汉工程大学学报,2026,48(04):449-457.[doi:10.19843/j.cnki.CN42-1779/TQ.202511008]
　CHEN Lihang,ZENG Xiangjin*,XIONG Ruifeng,et al. A multiscale adaptive feature image matching algorithm based on local feature transformer [J].Journal of Wuhan Institute of Technology,2026,48(04):449-457.[doi:10.19843/j.cnki.CN42-1779/TQ.202511008]
点击复制

基于局部特征变换器的多尺度自适应特征图像匹配算法
(/HTML)

《武汉工程大学学报》[ISSN:1674-2869/CN:42-1779/TQ]

卷:
48
期数:
2026年04期
页码:
449-457
栏目:
智能制造
出版日期:
2026-08-30

文章信息/Info

Title:
A multiscale adaptive feature image matching algorithm based on local feature transformer
文章编号:
1674 - 2869（2026）04 - 0449 - 09
作者:
陈礼航1; 曾祥进*1; 2; 熊瑞峰1; 段玉剑1
1. 武汉工程大学计算机科学与工程学院，湖北 武汉 430205;
2. 智能机器人湖北省重点实验室（武汉工程大学），湖北 武汉 430205
Author(s):
CHEN Lihang1; ZENG Xiangjin*1;  2; XIONG Ruifeng1; DUAN Yujian1
1. School of Computer Science and Engineering，Wuhan Institute of Technology，Wuhan 430205，China；
2. Hubei Provincial Key Laboratory of Intelligent Robotics（Wuhan Institute of Technology），Wuhan，430205，China
关键词:
图像匹配; 局部特征变换器; 特征重构; 特征描述; 自适应权重
Keywords:
image matching;  local feature transformer;  feature reconstruction;  feature descriptor;  adaptive weight
分类号:
TP391
DOI:
10.19843/j.cnki.CN42-1779/TQ.202511008
文献标志码:
A
摘要:
针对图像匹配神经网络提取的关键点特征，存在因视角变化导致匹配点对减少而匹配失败的问题，提出了一种基于局部特征变换器(LoFTR)的多尺度自适应特征图像匹配算法。通过改变提取的关键点特征组成结构，增强关键点的特征描述，强化对应点的相似性，从而优化特征匹配的效果。首先对残差网络（ResNet）和特征金字塔（FPN）输出的原始特征使用多尺度环形卷积（MRC）对齐关键点特征主方向；其次，设计自适应权重融合（AFW）计算环形卷积特征与具有全局位置信息特征的权重，按照权重进行融合，保留环形卷积特征的局部信息与全局特征位置信息的有效主成分；最后，引入边缘化样本一致性（MAGSAC++）算法，对匹配点对进一步筛选，减少噪声影响。在数据集Megadepth和Hpatches上验证所提算法，准确率达到87.62%，相较于LoFTR算法提升了1.78%，在阈值为5°、10°、20°条件下姿态估计误差的精确率-召回率曲线下面积（AUPRC）值均有提升，分别提高了0.8、1.2、1.3，结果证明本文算法有效提高了匹配点数量以及匹配质量，对猪只体长、体宽等参数测量的精度均有提升，为猪只养殖等生产活动提供更精确的测量。
Abstract:
To address the problem of matching failure caused by the reduction of matching point pairs due to viewpoint variations in the keypoint features extracted by the image matching neural network, in this study we proposed a multiscale adaptive feature image matching algorithm based on local feature transformer (LoFTR). By modifying the compositional structure of the extracted keypoint features, the method enhanced feature description and strengthened the similarity of corresponding points, thereby improving feature matching performance. First, multi-scale ring convolution (MRC) was applied to align the dominant orientations of keypoint features extracted from the raw features output by the residual network (ResNet) and feature pyramid network (FPN). Second, an adaptive feature weighting (AFW) module was designed to compute the fusion weights between the ring convolution features and features containing global positional information. These features were then fused according to the computed weights, preserving the effective principal components of both the local information from the ring convolution features and the global positional information. Finally the MAGSAC++ algorithm was introduced to further refine the matched point pairs to mitigate the impact of noise. The proposed algorithm model was validated on the MegaDepth and HPatches datasets, achieving an accuracy of 87.62%, which represented an improvement of 1.78% over the LoFTR algorithm. Under the thresholds of 5°, 10°, and 20°, the area under the precision-recall curve (AUPRC) of pose estimation increased by 0.8, 1.2, and 1.3, respectively. Results demonstrated that the proposed algorithm effectively increased the number of matching points and improved matching quality. It also enhanced the measurement accuracy of parameters such as swine body length and body width, thereby providing more precise measurements for production activities such as swine farming.

参考文献/References:

[ 1 ] 罗凯,徐俊武,杨敏.一种改进KNN的无人机图像快速拼接方法[J].武汉工程大学学报,2021,43(3):344-348.
[ 2 ] LOWE D G．Object recognition from local scale-invariant features[C]//Proceedings of the 7th IEEE International Conference on Computer Vision．Piscataway, NJ:IEEE，1999: 1150-1157．
[ 3 ] RUBLEE E, RABAUD V, KONOLIGE K, et al．ORB: an efficient alternative to SIFT or SURF[C]//2011 International Conference on Computer Vision. Piscataway, NJ: IEEE, 2011: 2564-2571.
[ 4 ] ROSTEN E, DRUMMOND T. Machine learning for high-speed corner detection[C]//Computer Vision-ECCV 2006. Berlin, Heidelberg: Springer, 2006: 430-443.
[ 5 ] CALONDER M, LEPETIT V, STRECHA C, et al. BRIEF: binary robust independent elementary features[C]// Computer Vision-ECCV 2010. Berlin，Heidelberg: Springer, 2010: 778-792.
[ 6 ] 罗云昊,黄战华,王康年.基于局部几何与纹理特征二进制描述的彩色点云配准算法[J].激光与光电子学进展，2026，63（2）：0215006.
[ 7 ] LIU Z, MAO H Z, WU C Y, et al. A ConvNet for the 2020s[C]//2022 IEEE/CVF Conference on computer vision and pattern recognition. Piscataway, NJ: IEEE, 2022: 11966-11976.
[ 8 ] DETONE D, MALISIEWICZ T, RABINOVICH A．Superpoint：self-supervised interest point detection and description[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops．Piscataway, NJ：IEEE, 2018: 337-349．
[ 9 ] SARLIN P E, DETONE D, MALISIEWICZ T, et al．SuperGlue: learning feature matching with graph neural networks[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition．Piscataway, NJ: IEEE, 2020: 4937-4946．
[10] LINDENBERGER P, SARLIN P E, POLLEFEYS M. LightGlue: local feature matching at light speed[C]//2023 IEEE/CVF International Conference on Computer Vision. Piscataway, NJ: IEEE 2023: 17627-17638.
[11] SUN J M, SHEN Z H, WANG Y A, et al．LoFTR: detector-free local feature matching with transformers[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition．Piscataway, NJ: IEEE, 2021: 8918-8927．
[12] LI X H, HAN K, LI S D, et al. Dual-resolution correspondence networks[C]// Advances in Neural Information Processing Systems. Red Hook, NY: Curran Associates, 2020, 33: 17346-17357.
[13] LIN T Y, DOLLAR P, GIRSHICK R, et al．Feature pyramid networks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition．Piscataway, NJ: IEEE,2017:936-944．
[14] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]//Advances in Neural Information Processing Systems. Red Hook, NY: Curran Associates 2017, 30.
[15] WANG L, ZHANG X Y, JIANG Z Q, et al. FMRT: learning accurate feature matching with reconciliatory transformer[J]. IEEE Transactions on Automation Science and Engineering, 2025, 22: 11826-11842.
[16] ZONG H L, YUAN X P, GAN S, et al. UAV image matching of mountainous terrain using the LoFTR deep learning model[J]. Frontiers in Earth Science. 2023，11:1203078.
[17] GAO Y, LIAO S H, DAI P S, et al. A general detector-free feature matching method for medical images[J]. Procedia Computer Science, 2025, 264: 80-93.
[18] BIAN H Q, CHEN Q F, ZHANG H L, et al. MambaLF: an efficient local feature extraction and matching with state space model[J]. Multimedia Systems, 2025, 31(4): 280.
[19] ZHAO X M, WU X M, MIAO J Y, et al. ALIKE: accurate and lightweight keypoint detection and descriptor extraction[J]. IEEE Transactions on Multimedia, 2022, 25: 3101-3112.
[20] POTJE G, CADAR F, ARAUJO A, et al. XFeat: accelerated features for lightweight image matching[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2024: 2682-2691.
[21] QU H Z, HU Z H, WU J Q. Multi-scale parallel gated local feature transformer[J]. Scientific Reports, 2025, 15: 7684.
[22] ZHENG C Y, LI S S, WANG C Y, et al. MSG: Robust multimodal remote sensing image matching using side window Gaussian space[J]. IEEE Transactions on Geoscience and Remote Sensing, 2025, 63: 4706223.
[23] BARATH D, NOSKOVA J, MATAS J. Marginali-zing sample consensus[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(11): 8420-8432.
[24] GIANG K T, SONG S, JO S. TopicFM: robust and interpretable topic-assisted feature matching[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Menlo Park, CA: AAAI Press, 2023, 37(2): 2447-2455.
[25] WANG Y F, HE X Y, PENG S D, et al. Efficient LoFTR: semi-dense local feature matching with sparse-like speed[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2024: 21666-21675.

相似文献/References:

备注/Memo

备注/Memo:
收稿日期：2025-11-21
基金项目：国家自然科学基金(61502354)；湖北省重大攻关项目（2025BEA005）
作者简介：陈礼航，硕士研究生。Email：2824055786@qq.com
*通信作者：曾祥进，博士，副教授。Email：xjzeng21@163.com
更新日期/Last Update: 2026-09-05