[1]李新卫,吴飞,荆晓远.基于协同矩阵分解的单标签跨模态检索[J].计算机技术与发展,2018,28(11):99-102.[doi:10.3969/ j. issn.1673-629X.2018.11.022]
　LI Xin-wei,WU Fei,JING Xiao-yuan.Cross-modality Retrieval Based on Collective Matrix Factorization with Single Label[J].,2018,28(11):99-102.[doi:10.3969/ j. issn.1673-629X.2018.11.022]
点击复制

基于协同矩阵分解的单标签跨模态检索()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
28
期数:
2018年11期
页码:
99-102
栏目:
智能、算法、系统工程
出版日期:
2018-11-10

文章信息/Info

Title:
Cross-modality Retrieval Based on Collective Matrix Factorization with Single Label
文章编号:
1673-629X(2018)11-0099-04
作者:
李新卫; 吴飞; 荆晓远
南京邮电大学 自动化学院,江苏 南京 210023
Author(s):
LI Xin-wei; WU Fei; JING Xiao-yuan
School of Automation,Nanjing University of Posts and Telecommunications,Nanjing 210023,China
关键词:
协同矩阵分解; 哈希函数; 图正则化; 稀疏图; 跨模态检索
Keywords:
collective matrix factorization; hash function; graph regularization; sparse graph; cross-modality retrieval
分类号:
TP181
DOI:
10.3969/ j. issn.1673-629X.2018.11.022
文献标志码:
A
摘要:
针对跨模态检索存在的存储空间大、检索速度慢等缺点,提出了一种基于协同矩阵分解单标签跨模态检索方法,目标函数主要由协同矩阵分解、哈希函数和保持局部流形几何结构的图正则化三部分组成。 矩阵分解学习训练数据集在低维潜在语义空间的哈希编码的简洁表示;哈希函数用来学习投影,将训练集外的样本表示成学习到的子空间的哈希码, 根据汉明排序进行相似性搜索;图正则化用来保持原始空间的局部流行几何结构,该算法将这三部分有机地结合起来。为了证实该算法的有效性,在两个常用的数据集 Wiki 和 Pascal VOC 2007 进行了大量的实验,并与一些常用的相关方法进行了比较,结果证明了该算法的优越性。
Abstract:
Aiming at the disadvantages of large storage space and slow retrieval speed,we propose a novel cross-modality retrieval algorithm based on collective matrix factorization with single label. The objective function is mainly composed of cooperative matrix factorization,hash function and graph regularization of local manifold geometry structure. In particular,matrix decomposition learns the concise representation of hash coding of training data sets in low-dimensional latent semantic space. The hash function is used to learn the projection,the samples outside the training set are represented as the hash code of the learned subspace,and the similarity search is conducted according to Hamming sequence. Graph regularization is used to maintain the local popular geometry of the original space. We conjecture all these to improve the retrieval accuracy. Experiment on two cross-modality visual search datasets,Wiki and Pascal VOC 2007,shows that the proposed algorithm can significantly outperform the various state-of-the-art relevant methods.

相似文献/References:

[1]钱萌 程玉胜 程树林 叶敏 汪智华 齐乐.基于分治策略求解方程根的个数[J].计算机技术与发展,2006,(09):41.
　QIAN Meng,CHENG Yu-sheng,CHENG Shu-lin,et al.Seeking for Roots of Equation Based on Strategy of Divide and Conquer[J].,2006,(11):41.
[2]祝彦斌,王春玲.一种Hash特征隐藏的加盐信息摘要模型[J].计算机技术与发展,2013,(03):134.
　ZHU Yan-bin,WANG Chun-ling.A Message-digest Model with Salt and Hidden Feature of Hash[J].,2013,(11):134.
[3]马丽,窦家维,吴艳梅. 具有不可关联性的承诺方案[J].计算机技术与发展,2017,27(05):108.
　MA Li,DOU Jia-wei,WU Yan-mei. Non-malleable Commitment Schemes[J].,2017,27(11):108.

更新日期/Last Update: 2018-11-10