基于哈希的大规模多标签图像搜索方法研究-计算机技术专业论文.docxVIP

山东大学硕士学位论文3．2．1 山东大学硕士学位论文 3．2．1 GIST 22 ：；．2．2 SIFT ．23 3．3用标签信息来投影 ．23 3．4用PCA来投影 25 3．5用ITQ学习旋转矩阵 27 3．6 Sgn函数得到哈希码 28 3．7方法总结 ．28 第四章实验设计 ．30 4．1数据集 ．30 4．1．1 NUS．WIDE ．30 4．1．2 CIFAR．1 OO ．．3 1 4．2比较方法 31 4．3平衡因子 ．32 4．4度量方法 ．32 4．4．1 MeanAverage Precision(MAP) ．．32 4．4．2 Precision、Recall ．33 4．4．3 Precision within hamming distance 2 ．34 4．5如何定义good neighbors 34 4．6实验结果 ．35 第五章结论与展望 39 5．1结论 39 5．2展望 ．39 参敬献 致 谢 43 攻读学位期间发表的学术论文目录 ．44 万方数据 山东大学硕士学位论文CoNTENTS 山东大学硕士学位论文 CoNTENTS Chinese Abstract．．．．．．． ． ．． ． ． ． I English Abstract ．．III Chapter 1 Introduction ．．．． ．．． ． ．． ． ．． ．． ．．1 1．1 Research background 1 1．2 Research ofimage search ．． ． ． ． ． ． ．． ． ． ．．．1 1．2．1 TbXt based Image Retrieval ．．2 1．2．1 Context based Image Retrieval． ．． ． ．． ． ． ． ． ． ． ． ．．4 1．2．1 Comparison ofTBIR and CBIR ．5 1．3 The main works ofthis paper ．7 1．4 The structure ofthis paper 7 Chapter 2 Related work ．． ． ． ． ． ．． ．9 2．1 Briefintroduction ofhash method ．9 2．1．1 What is learning to hash ．9 2．1．2 Development trend oflearning to hash 10 2．2 Hashing based Image Retrieval 12 2．3 Unsupervised hashing methods ． ． ．14 2．3．1 Local sensitive hashing ． 14 2．3．2 Spectral hashing 15 2．4 Supervised hashing metllod 16 2．4．1 RBMs ．1 6 2．5 Semi—supervised hashing method 1 7 2．6 Multi—label dimension reduction technology ．1 8 2．6．1 Canonical correlation analysis ．．19 Chapter 3 Specific scheme of MLSH ． ．．21 3．1 Problem description 2l 3．2 Feature extraction method ． 22 万方数据 山东大学硕士学位论文3．2．1 山东大学硕士学位论文 3．2．1 GIST ．22 3．2．2 SIFT ．．23 3．3 Projection using the label information ． ． ．． ． ． ．． ． ．． ． ．23 3．4 Projection using PCA ． ．．25 3．5 Learning using ITQ rotation matrix ．．27 3．6 Getting hash code using sgn function 28 3．6 Method summary ．28 Chapter 4 Experiments design ．． ． ． ． · 30 4．1 Data sets 30 4．1．1 NUS．WIDE 。30 4．1．2 CIFAR-100 31 4．2 Comparison methods ．3 1 4．3 Balance factor 32 4．4 Measurement method 32 4．4．1 Mean Average Precision(MAP)