基于深度学习耦合稀疏语义度量的商标检索算法
|
摘要
目的针对当前商标图像检索中的语义鸿沟问题,提出一种深度学习耦合稀疏语义度量的商标图像检索方案,有效抑制噪声干扰,降低冗余特征维数。方法首先,根据由卷积与池化组成的无监督学习机制,对输入商标图像进行多层特征提取,输出一维特征向量。随后,通过L2-支持向量机(L2-SVM)进行分类,利用特征向量进行训练,获得多级联特征。然后,根据商标图像的多级联特征和用户标签信息的异构数据结构,设计一种稀疏语义度量方法进行相似检索,减少语义鸿沟。此外,引入一种混合范数作为相似度量的稀疏约束,以抑制原始输入空间中的冗余特征维数和噪声,优化检索结果。结果实验表明,与当前流行的商标检索方案相比,所提算法具有更高的检索精度,其输出的结果中仅有1幅无关图像。结论该方案具有较高的检索精度和较强的鲁棒性,在商标检测、商标保护等方面中具有良好的应用价值。
The work aims to propose a trademark image retrieval scheme with deep learning coupled sparse semantic measure to effectively restrain noise interference and reduce redundant feature dimension, with respect to the problem of semantic gap in trademark image retrieval. Firstly, according to the unsupervised learning mechanism composed of convolution and cistern, the multi-layer feature extraction of input trademark image were carried out to output the one-dimensional feature vector. Then, the L2-support vector machine(L2-SVM) was used to classify the feature vectors, and the multilevel features were obtained based on the training with feature vectors. Then, according to the multilevel feature of the trademark image and the heterogeneous data structure of the user's label information, a sparse semantic measure method was designed for similar retrieval, which reduced the semantic gap. In addition, a mixed norm was introduced as a sparse constraint of similarity measure to suppress the redundant feature dimension and noise in the original input space and optimize the retrieval results. The experiment showed that, compared with the current popular trademark retrieval scheme, the proposed algorithm had higher retrieval accuracy, whose output results only had one irrelevant image. The proposed scheme has higher retrieval precision and stronger robustness, and it has good application value in trademark detection, trademark protection and so on.
The work aims to propose a trademark image retrieval scheme with deep learning coupled sparse semantic measure to effectively restrain noise interference and reduce redundant feature dimension, with respect to the problem of semantic gap in trademark image retrieval. Firstly, according to the unsupervised learning mechanism composed of convolution and cistern, the multi-layer feature extraction of input trademark image were carried out to output the one-dimensional feature vector. Then, the L2-support vector machine(L2-SVM) was used to classify the feature vectors, and the multilevel features were obtained based on the training with feature vectors. Then, according to the multilevel feature of the trademark image and the heterogeneous data structure of the user's label information, a sparse semantic measure method was designed for similar retrieval, which reduced the semantic gap. In addition, a mixed norm was introduced as a sparse constraint of similarity measure to suppress the redundant feature dimension and noise in the original input space and optimize the retrieval results. The experiment showed that, compared with the current popular trademark retrieval scheme, the proposed algorithm had higher retrieval accuracy, whose output results only had one irrelevant image. The proposed scheme has higher retrieval precision and stronger robustness, and it has good application value in trademark detection, trademark protection and so on.
引文
[1]GIUSEPPE A,FABRIZO F,LUCIA V.Aggregating Binary Local Descriptors for Image Retrieval[J].Multimedia Tools and Applications,2018,77(5):5385-5415.
[2]黄友文,黄雅兰.一种基于Harris与SIFT算子结合的商标搜索方法[J].电视技术,2014,38(3):35-38.HUANG You-wen,HUANG Ya-lan.Method of Trademark Search Based on Harris and SIFT[J].Video Engineering,2014,38(3):35-38.
[3]LU Wei-ding,HENG Jiang.A Document Expansion Framework for Tag-based Image Retrieval[J].Aslib Journal of Information Management,2018,70(1):47-65.
[4]刘剑英.基于角点描述和区域特征的商标图像检索[D].西安:西安电子科技大学,2017.LIU Jian-ying.Trademark Image Retrieval Based on Corner Description and Regional Feature[D].Xi'an:Xidian University,2017.
[5]AKRITI N.Trademark Image Retrieval Using Weighted Combination of Sift and HSV Correlogram[J].International Journal of Computer Applications in Technology,2016,54(1):61-67.
[6]YAN Yi-jun,REN Jin-chang,LI Yin-sheng.Adaptive Fusion of Color and Spatial Features for Noise-robust Retrieval of Colored Logo and Trademark Images[J].Multidimensional Systems and Signal Processing,2016,27(4):945-968.
[7]王伟凝,王励,赵明权.基于并行深度卷积神经网络的图像美感分类[J].自动化学报,2016,42(6):904-914.WANG Wei-ning,WANG Li,ZHAO Ming-quan.Image Aesthetic Classification Using Parallel Deep Convolution Neural Networks[J].Acta Automatic Sinica,2016,42(6):904-914.
[8]罗富贵,李明珍.基于卷积核分解的深度CNN模型结构优化及其在小图像识别中的应用[J].井冈山大学学报(自然科学版),2018,39(2):31-38.LUO Fu-gui,LI Ming-zhen.Structural Optimization of Deep CNN Model Based on Convolution Kernel Decomposition and Its application in Small Image Recognition[J].Journal of Jinggangshan University(Natural Science Edition),2018,39(2):31-38.
[9]房卫东,石志东,单联海.一种基于BETA分布抗On-off攻击的信任机制[J].系统仿真学报,2015,27(11):2722-2728.FANG Wei-dong,SHI Zhi-dong,SHAN Lian-hai.Trusted Scheme for Defending On-Off Attack Based on BETA Distribution[J].Journal of System Simulation,2015,27(11):2722-2728.
[10]王维智.基于特征提取和特征选择的级联深度学习模型研究[D].哈尔滨:哈尔滨工业大学,2015.WANG Wei-zhi.Research on Cascaded Deep Learning Model Based on Feature Extraction and Feature Selection[D].Harbin:Harbin Institute of Technology,2015.
[11]LIU Jing,LI Ze-chao,LU Han-qing.Sparse Semantic Metric Learning for Image Retrieval[J].Multimedia Systems,2014,20(6):635-643.
[12]ESKANDAR N.Existence and Convergence Theorems for Bregman Best Proximity Points in Reflexive Banach Spaces[J].Journal of Fixed Point Theory and Applications,2018,20(1):453-468.
[13]郭强.基于JBLD度量的快速记忆梯度距离度量方法[J].光电技术应用,2016,31(5):51-54.GUO Qiang.Fast Memory Gradient Algorithm Based on Jensen-Bregman LogDet Metric[J].Electro-optic Technology Application,2016,31(5):51-54.
[14]李瑞雪,夏斌,袁文浩.基于多元变量Taylor级数展开模型的定位算法[J].计算机应用研究,2016,33(6):1853-1856.LI Rui-xue,XIA Bin,YUAN Wen-hao.Localization Algorithm Based on Multivariable Taylor Series Expansion Model[J].Application Research of Computers,2016,33(6):1853-1856.
[15]柯圣财,赵永威,李弼程.基于卷积神经网络和监督核哈希的图像检索方法[J].电子学报,2017,45(1):157-163.KE Sheng-cai,ZHAO Yong-wei,LI Bi-cheng.Image Retrieval Based on Convolutional Neural Network and Kernel-Based Supervised Hashing[J].Acta Electronica Sinica,2017,45(1):157-163.
[16]相子喜,吕学强,张凯.基于有向图模型的多模态新闻图像检索研究[J].科学技术与工程,2016,16(3):78-84.XIANG Zi-xi,LYU Xue-qiang,ZHANG Kai.Research on Multimodal News Image Retrieval Based on Directed Graph Model[J].Science Technology and Engineering,2016,16(3):78-84.
[2]黄友文,黄雅兰.一种基于Harris与SIFT算子结合的商标搜索方法[J].电视技术,2014,38(3):35-38.HUANG You-wen,HUANG Ya-lan.Method of Trademark Search Based on Harris and SIFT[J].Video Engineering,2014,38(3):35-38.
[3]LU Wei-ding,HENG Jiang.A Document Expansion Framework for Tag-based Image Retrieval[J].Aslib Journal of Information Management,2018,70(1):47-65.
[4]刘剑英.基于角点描述和区域特征的商标图像检索[D].西安:西安电子科技大学,2017.LIU Jian-ying.Trademark Image Retrieval Based on Corner Description and Regional Feature[D].Xi'an:Xidian University,2017.
[5]AKRITI N.Trademark Image Retrieval Using Weighted Combination of Sift and HSV Correlogram[J].International Journal of Computer Applications in Technology,2016,54(1):61-67.
[6]YAN Yi-jun,REN Jin-chang,LI Yin-sheng.Adaptive Fusion of Color and Spatial Features for Noise-robust Retrieval of Colored Logo and Trademark Images[J].Multidimensional Systems and Signal Processing,2016,27(4):945-968.
[7]王伟凝,王励,赵明权.基于并行深度卷积神经网络的图像美感分类[J].自动化学报,2016,42(6):904-914.WANG Wei-ning,WANG Li,ZHAO Ming-quan.Image Aesthetic Classification Using Parallel Deep Convolution Neural Networks[J].Acta Automatic Sinica,2016,42(6):904-914.
[8]罗富贵,李明珍.基于卷积核分解的深度CNN模型结构优化及其在小图像识别中的应用[J].井冈山大学学报(自然科学版),2018,39(2):31-38.LUO Fu-gui,LI Ming-zhen.Structural Optimization of Deep CNN Model Based on Convolution Kernel Decomposition and Its application in Small Image Recognition[J].Journal of Jinggangshan University(Natural Science Edition),2018,39(2):31-38.
[9]房卫东,石志东,单联海.一种基于BETA分布抗On-off攻击的信任机制[J].系统仿真学报,2015,27(11):2722-2728.FANG Wei-dong,SHI Zhi-dong,SHAN Lian-hai.Trusted Scheme for Defending On-Off Attack Based on BETA Distribution[J].Journal of System Simulation,2015,27(11):2722-2728.
[10]王维智.基于特征提取和特征选择的级联深度学习模型研究[D].哈尔滨:哈尔滨工业大学,2015.WANG Wei-zhi.Research on Cascaded Deep Learning Model Based on Feature Extraction and Feature Selection[D].Harbin:Harbin Institute of Technology,2015.
[11]LIU Jing,LI Ze-chao,LU Han-qing.Sparse Semantic Metric Learning for Image Retrieval[J].Multimedia Systems,2014,20(6):635-643.
[12]ESKANDAR N.Existence and Convergence Theorems for Bregman Best Proximity Points in Reflexive Banach Spaces[J].Journal of Fixed Point Theory and Applications,2018,20(1):453-468.
[13]郭强.基于JBLD度量的快速记忆梯度距离度量方法[J].光电技术应用,2016,31(5):51-54.GUO Qiang.Fast Memory Gradient Algorithm Based on Jensen-Bregman LogDet Metric[J].Electro-optic Technology Application,2016,31(5):51-54.
[14]李瑞雪,夏斌,袁文浩.基于多元变量Taylor级数展开模型的定位算法[J].计算机应用研究,2016,33(6):1853-1856.LI Rui-xue,XIA Bin,YUAN Wen-hao.Localization Algorithm Based on Multivariable Taylor Series Expansion Model[J].Application Research of Computers,2016,33(6):1853-1856.
[15]柯圣财,赵永威,李弼程.基于卷积神经网络和监督核哈希的图像检索方法[J].电子学报,2017,45(1):157-163.KE Sheng-cai,ZHAO Yong-wei,LI Bi-cheng.Image Retrieval Based on Convolutional Neural Network and Kernel-Based Supervised Hashing[J].Acta Electronica Sinica,2017,45(1):157-163.
[16]相子喜,吕学强,张凯.基于有向图模型的多模态新闻图像检索研究[J].科学技术与工程,2016,16(3):78-84.XIANG Zi-xi,LYU Xue-qiang,ZHANG Kai.Research on Multimodal News Image Retrieval Based on Directed Graph Model[J].Science Technology and Engineering,2016,16(3):78-84.