[1]王立新,江加和.基于深度学习的显著性区域的图像检索研究[J].应用科技,2018,45(06):63-67.[doi:10.11991/yykj.201803012]
 WANG Lixin,JIANG Jiahe.Research on image retrieval of saliency region based on deep learning[J].Applied science and technology,2018,45(06):63-67.[doi:10.11991/yykj.201803012]
点击复制

基于深度学习的显著性区域的图像检索研究(/HTML)
分享到:

《应用科技》[ISSN:1009-671X/CN:23-1191/U]

卷:
第45卷
期数:
2018年06期
页码:
63-67
栏目:
计算机技术与应用
出版日期:
2018-11-05

文章信息/Info

Title:
Research on image retrieval of saliency region based on deep learning
作者:
王立新 江加和
北京航空航天大学 自动化科学与电气工程学院, 北京 100191
Author(s):
WANG Lixin JIANG Jiahe
School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, China
关键词:
图像检索卷积神经网络局部特征全局特征显著性区域相似度深度学习模型训练
Keywords:
image retrievalconvolutional neural networklocal featureglobal featuresaliency regionsimilaritydeep learningmodel training
分类号:
TP391
DOI:
10.11991/yykj.201803012
文献标志码:
A
摘要:
利用神经网络提取的图像全局特征包含图像上的冗余信息,影响检索的精度,为了解决这个问题,提出了一种基于VGG16的改进网络结构、保留图像空间信息、提取图像显著性区域局部特征的算法。首先利用改进的网络对数据进行训练,得到准确率较高的模型,利用训练好的模型对所有图像使用类激活映射(CAM)的方法定位出图像的显著性区域;然后利用相同的模型提取局部显著性区域特征,构建图像数据库;最后对查询图像使用距离比较函数(L2)计算相似度,按相似度大小排列返回相似图像。在Corel数据集上,对比提取神经网络全局特征以及使用传统SIFT特征构建的K-means模型,使用局部显著性区域特征有较高的检索精度。实验结果表明,该模型有较好的检索效果。
Abstract:
At present, the global features of images extracted by neural networks contain redundant information on the images, which affects the accuracy of retrieval. In order to solve this problem, an improved network structure based on VGG16 is proposed, which preserves the spatial information of images and extracts the local features of image salience. The algorithm first uses the improved network to train the data to obtain a model with higher accuracy, uses the trained model to locate significant region of the image by class activation mapping (CAM); then uses the trained model mentioned above to extract local features of image salience, and then constructs an image database on this basis; at last calculates similarity by using a distance comparison function (L2) on the query image, returning similar images that are sequenced according to similarity. On the Corel dataset, comparing the K-means models that extract global features of neural network and that are constructed using the scale invariant feature transform (SIFT) features, the method using local features of image salience has higher retrieval accuracy. The experimental results show that the model has a good search effect.

参考文献/References:

[1] 杭燕, 杨育彬, 陈兆乾. 基于内容的图像检索综述[J]. 计算机应用研究, 2002, 19(9):9-13, 29.
[2] LOWE D G. Distinctive image features from scale-invariant keypoints[J]. International journal of computer vision, 2004, 60(2):91-110.
[3] HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786):504-507.
[4] LAI Hanjiang, PAN Yan, LIU Ye, et al. Simultaneous feature learning and hash coding with deep neural networks[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, USA, 2015, 3270-3278.
[5] DENG Jia, BERG A C, LI Feifei. Hierarchical semantic indexing for large scale image retrieval[C]//CVPR 2011. Colorado Springs, USA, 2011.
[6] BABENKO A, SLESAREV A, CHIGORIN A, et al. Neural codes for image retrieval[M]//FLEET D, PAJDLA T, SCHIELE B, et al. Computer Vision-ECCV 2014. Cham, Germany:Springer, 2014.
[7] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition.[2017-12-20]. https://arxiv.org/abs/1409.1556.
[8] ZHOU Bolei, BAU D, OLIVA A, et al. Interpreting deep visual representations via network dissection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018(99):1-14.
[9] LIN Min, CHEN Qiang, YAN Shuicheng. Network in network.[2017-12-20]. https://arxiv.org/abs/1312.4400.
[10] IOFFE S, SZEGEDY C. Batch normalization:accelerating deep network training by reducing internal covariate shift.[2017-12-20]. https://arxiv.org/pdf/1502.03167v3.
[11] ZHOU Bolei, KHOSLA A, LAPEDRIZA A, et al. Learning deep features for discriminative localization[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, USA, 2016.
[12] DENG Jia, BERG A C, LI Feifei. Hierarchical semantic indexing for large scale image retrieval[C]//Proceedings of CVPR 2011. Colorado Springs, USA, 2011.
[13] HE Kaiming, WEN Fang, SUN Jian. K-means hashing:An affinity-preserving quantization method for learning binary compact codes[C]//2013 IEEE Conference on Computer Vision and Pattern Recognition. Portland, USA, 2013:2938-2945.
[14] BABENKO A, SLESARE A, CHIGORIN A, et al. Neural codes for image retrieval[C]//13th European Conference Computer Vision. Zurich, Switzerland, 2014.

相似文献/References:

[1]宋佳蓉,杨忠,张天翼,等.基于卷积神经网络和多类SVM的交通标志识别[J].应用科技,2018,45(05):71.[doi:10.11991/yykj.201710009]
 SONG Jiarong,YANG Zhong,ZHANG Tianyi,et al.Traffic signs recogniton based on convolutional neural networks and multi-class SVM[J].Applied science and technology,2018,45(06):71.[doi:10.11991/yykj.201710009]
[2]郜雨桐,宁慧,王巍,等.基于卷积神经网络的车辆型号识别研究[J].应用科技,2018,45(06):53.[doi:10.11991/yykj.201803011]
 GAO Yutong,NING Hui,WANG Wei,et al.Research on vehicle model identification based on convolutional neural network[J].Applied science and technology,2018,45(06):53.[doi:10.11991/yykj.201803011]

备注/Memo

备注/Memo:
收稿日期:2018-03-21。
基金项目:国家自然科学基金项目(61673038)
作者简介:王立新(1993-),男,硕士研究生;江加和,男,副教授,博士
通讯作者:王立新,E-mail:13269535100@sina.cn
更新日期/Last Update: 2018-11-02