摘要:在信息技术飞速进步的时代,各种图像获取设备也随之普及,大量的图像数据被收集。如何管理与检索大量的图像数据是值得研究的问题。本文针对跨模态的哈希检索背景进行简述,对国内外现状做出综述,主要研究相关的三种算法:典型相关分析(CCA)、正交投影学习语义关联最大化哈希(SCM-Orth)和非正交投影连续学习语义关联最大化哈希(SCM-Seq)[1]。文中对上述算法进行介绍与实现,在多个数据集上进行实验进行算法对比,在此基础上进行算法改进与实现,并给出改进算法的思路与详细推导。29837
毕业论文关键词:跨模态;哈希检索;典型相关分析;语义最大化哈希
Study of cross-modal hashing with Semantic Correlation Maximization retrieval
Abstract:For the rapid development of information technology in this time, various image acquisition devices become popular. A large number of images are collected today. How to manage and retrieve a large amount of image data is a problem worthy of study. This article in view of the cross-modal hashing retrieval background briefly, then make a review about the current situation and development. Main research relates to three algorithms: Canonical Correlation Analysis(CCA), Semantic Correlation Maximization Learning for Orthogonal Projection(SCM-Orth), Semantic Correlation Maximization Sequential Learning for Non-Orthogonal Projection(SCM-Seq). In this article, the above algorithm is introduced and implemented, the algorithms are compared on several data sets, and on this basis, the algorithms are improved and realized.
Key words: cross-modal; hashing; Canonical Correlation Analysis; hashing with Semantic Correlation Maximization
目 录
摘要 1
关键词 1
1.绪论 1
1.1 研究意义 1
1.2 国内外研究现状 2
1.3 研究目的与内容 4
1.4 论文组织结构 4
2 算法研究预备知识 4
2.1 拉格朗日乘子法 4
2.2 矩阵的迹的性质及求导法则 5
3 算法阐述实现 5
3.1 典型相关分析(CCA) 5
3.2 语义最大化哈希(SCM)方法标记与问题定义 7
3.3 正交投影学习的语义关联最大化(SCM-Orth)[1] 8
3.3.1 模型构建 8
3.3.2 算法简述 8
3.4 非正交投影连续学习的语义关联最大化(SCM-Seq)[1] 9
4 基于CCA和SCM的算法改进 10
4.1 算法改进思路 10
4.2 SCM-Orth算法修改为SCM-Orth-dCCA的详细步骤与推导 11
4.3 SCM-Seq算法改进为SCM-Seq-dCCA简述 13
5 训练算法实现流程 13
5.1 典型相关分析(CCA) 13
5.2 正交投影学习的语义关联最大化哈希(SCM-Orth) 13
5.3 非正交投影连续学习的语义关联最大化哈希(SCM-Seq) 14
5.4 SCM-Orth改进算法SCM-Orth-dCCA 15
5.5 SCM-Seq改进算法SCM-Seq-dCCA 15
6 实验与分析对比 16
6.1 数据集来源 16
6.1.1 Wiki Data-set数据集 16
6.1.2 NUS-WIDE数据集 16
6.2 数据集选取 16
6.3 算法测试评价标准 17 语义关联最大化的跨模态哈希检索算法研究+源代码:http://www.youerw.com/jisuanji/lunwen_25243.html