Enhanced Feature Descriptor Based on Attention Mechanism

doi:10.19678/j.issn.1000-3428.0058788

Abstract

Abstract: When the traditional manual method is used to obtain the feature descriptors,it can not guarantee the correct matching of the feature points in the nonlinear deformation state,and can not solve the problem of feature description when the image has large deformation.To solve the problem of feature point matching in MR images of flexible biological tissues,an enhancement descriptor combined with attention mechanism is proposed.The multi-layer perceptron is used to encode the position information of feature points and combine it with the initial descriptor of feature points.The method of self attention and cross attention in graph attention neural network is combined,and the node information is transmitted by message passing method. By making full use of the hierarchy of graph attention neural network,the node information between different levels is fused and the feature descriptor is obtained.The experimental results on real flexible tissue MR image data set show that the describe performance of this descriptor is better than SIFT,SURF,DAISY and GIH descriptors,and it is suitable for real MR image matching.

Key words: attention mechanism, MR image, flexible biological tissues, deformation, descriptor

摘要： 传统手工获取特征描述子的方式不能保证特征点在非线性形变状态下进行正确匹配，无法有效解决图像存在较大形变时的特征描述问题。针对柔性生物组织MR影像特征点匹配问题，提出一种结合注意力机制的增强描述子。采用多层感知机对特征点位置信息进行编码并与特征点的初始描述子相结合，将图注意力神经网络中的自我注意与交叉注意的方法相结合，并运用消息传递方法传递节点信息。通过充分利用图注意力神经网络的层次性，以融合不同层次间的节点信息并最终获得特征描述子。在真实柔性生物组织MR影像数据集上的实验结果表明，该描述子相比SIFT、SURF、DAISY、GIH描述子的描述性能更优，且适用于真实MR影像的匹配任务。

关键词: 注意力机制, MR图像, 柔性生物组织, 形变, 描述子

CLC Number:

TP301.6

CHEN Jia, HU Haobo, HE Ruhan, HU Xinrong. Enhanced Feature Descriptor Based on Attention Mechanism[J]. Computer Engineering, 2021, 47(5): 260-266.

陈佳, 胡浩博, 何儒汉, 胡新荣. 基于注意力机制的增强特征描述子[J]. 计算机工程, 2021, 47(5): 260-266.

/ / Recommend / Download Citations

URL: http://www.ecice06.com/EN/10.19678/j.issn.1000-3428.0058788

http://www.ecice06.com/EN/Y2021/V47/I5/260

References

[1] HU Lichao,SHI Zaifeng,PANG Ke,et al.Improved Harris feature point detection algorithm for image matching[J].Computer Engineering,2015,41(10):216-220.(in Chinese)扈立超,史再峰,庞科,等.用于图像匹配的改进Harris特征点检测算法[J].计算机工程,2015,41(10):216-220.
[2] BO Lianbin,DIAO Haifeng,SUN Dedi,et al.Multi-modality medical image registration based on gradient generalized nearest-neighbor graph[J].Computer Engineering,2012,38(10):200-202.(in Chinese)卜令斌,赵海峰,孙登第,等.基于梯度广义近邻图的多模医学图像配准[J].计算机工程,2012,38(10):200-202.
[3] HATTIS C G,STEPHENS M J.A combined corner and edge detector[EB/OL].[2020-05-20].https://max.book118.com/html/2017/0614/114991147.shtm.
[4] MIKOLAJCZYK K,SCHMID C.A performance evaluation of local descriptors[J].IEEE Transaction on Pattern Analysis and Machine Intelligence,2005,27(10):1615-1630.
[5] LU Lai,WANG Junmin,FAN Rui.Binary descriptor for images based on Adaboost[J].Computer Engineering,2016,42(6):230-234,240.(in Chinese)卢来,王军民,范锐.基于自适应增强的图像二值描述子[J].计算机工程,2016,42(6):230-234,240.
[6] LOWE D G.Distinctive image features from scale invariant keypoints[J].International Journal of Computer Vision,2004,60(2):91-110.
[7] YIN Chunxia,XU De,LI Chengrong,et al.SIFT feature detection and matching based on salient map[J].Computer Engineering,2012,38(16):189-191.(in Chinese)尹春霞,徐德,李成荣,等.基于显著图的SIFT特征检测与匹配[J].计算机工程,2012,38(16):189-191.
[8] CHENG Dezhi,LI Yanjun,YU Ruixing.Image matching method based on improved SIFT algorithm[J].Computer Simulation,2011,28(7):285-289.(in Chinese)程德志,李言俊,余瑞星.基于改进SIFT算法的图像匹配方法[J].计算机仿真,2011,28(7):285-289.
[9] RUBLEE E,RABAUD V,KONOLIGE K,et al.ORB:an efficient alternative to SIFT or SURF[C]//Proceedings of International Conference on Computer Vision.Washington D.C.,USA:IEEE Press,2012:1-8.
[10] WU Zonghan,PAN Shirui,CHEN Fengwen,et al.A comprehensive survey on graph neural networks[J].IEEE Transactions on Neural Networks and Learning Systems,2020,99:1-21.
[11] ROSTEN E,DRUMMOND T.Machine learning for high-speed corner detection[C]//Proceedings of the 9th European Conference on Computer Vision.Graz,Austria:[s.n.],2006:430-443.
[12] ROSTEN E,DRUMMOND T.Fusing points and lines for high performance tracking[C]//Proceedings of ICCV'05.Washington D.C.,USA:IEEE Press,2005:1508-1511.
[13] GUAN Shiyong,LIU Lizhong,YAN Bin,et al.An image feature descriptor based on stable region[J].Computer Engineering,2012,38(18):186-189.(in Chinese)管士勇,陆利忠,闫镔,等.一种基于稳定区域的图像特征描述子[J].计算机工程,2012,38(18):186-189.
[14] BAY H,TUYTELAARS T,VAN G L.SURF:speeded up robust features[C]//Proceedings of the 9th European Conference on Computer Vision.Graz,Austria:[s.n.],2006:404-417.
[15] DETONE D,MALISIEWICZ T,RABINOVOCH A.Superpoint:self-supervised interest point detection and description[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops.Washington D.C.,USA:IEEE Press,2018:1-12.
[16] LU Yi,CHEN Yaran,ZHAO Dongbin,et al.Graph-FCN for image semantic segmentation[J].Advances in Neural Networks,2019,11554:97-105.
[17] FAN Weiqi,MA Yao,LI Qing,et al.Graph neural networks for social recommendation[C]//Proceedings of the World Wide Web Conference.New York,USA:ACM Press,2019:417-426.
[18] XU Sheng,WANG Tishuang,LI Peifeng,et al.Multi-layer attention network based on Chinese implicit discourse relation recognition[J].Journal of Chinese Information Processing,2019,33(8):12-19,35.(in Chinese)徐昇,王体爽,李培峰,等.运用多层注意力神经网络识别中文隐式篇章关系[J].中文信息学报,2019,33(8):12-19,35.
[19] NI Tongguang.A study on classification and feature selection based on transfer learning and its application[D].Wuxi:Jiangnan University,2015.(in Chinese)倪彤光.基于迁移学习的特征选择与分类方法及其应用研究[D].无锡:江南大学,2015.
[20] TOLA E,LEPETIT V,FUA P.DAISY:an efficient dense descriptor applied to wide-baseline stereo[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2010,32(5):815-830.
[21] LIANG H,JACOBS D W.Deformation invariant image matching[C]//Proceedings of the 10th IEEE International Conference on Computer Vision.Washington D.C.,USA:IEEE Press,2005:1-9.

Please choose a citation manager

Content to export