现有的深度哈希图像检索方法主要采用卷积神经网络,提取的深度特征的相似性表征能力不足.此外,三元组深度哈希主要从小批量数据中构建局部三元组样本,样本数量较少,数据分布缺失全局性,使网络训练不够充分且收敛困难.针对上述问题,文中...现有的深度哈希图像检索方法主要采用卷积神经网络,提取的深度特征的相似性表征能力不足.此外,三元组深度哈希主要从小批量数据中构建局部三元组样本,样本数量较少,数据分布缺失全局性,使网络训练不够充分且收敛困难.针对上述问题,文中提出基于类相似特征扩充与中心三元组损失的哈希图像检索模型(Hash Image Retrieval Based on Category Similarity Feature Expansion and Center Triplet Loss,HRFT-Net).设计基于Vision Transformer的哈希特征提取模块(Hash Feature Extraction Module Based on Vision Transformer,HViT),利用Vision Transformer提取表征能力更强的全局特征信息.为了扩充小批量训练样本的数据量,提出基于类约束的相似特征扩充模块(Similar Feature Expansion Based on Category Constraint,SFEC),利用同类样本间的相似性生成新特征,丰富三元组训练样本.为了增强三元组损失的全局性,提出基于Hadamard的中心三元组损失函数(Central Triplet Loss Function Based on Hadamard,CTLH),利用Hadamard为每个类建立全局哈希中心约束,通过增添局部约束与全局中心约束的中心三元组加速网络的学习和收敛,提高图像检索的精度.在CIFAR10、NUS-WIDE数据集上的实验表明,HRFT-Net在不同长度比特位哈希码检索上的平均精度均值较优,由此验证HRFT-Net的有效性.展开更多
It is well known that robustness, fragility, and security are three important criteria of image hashing; however how to build a system that can strongly meet these three criteria is still a challenge. In this paper, a...It is well known that robustness, fragility, and security are three important criteria of image hashing; however how to build a system that can strongly meet these three criteria is still a challenge. In this paper, a content-based image hashing scheme using wave atoms is proposed, which satisfies the above criteria. Compared with traditional transforms like wavelet transform and discrete cosine transform (DCT), wave atom transform is adopted for the sparser expansion and better characteristics of texture feature extraction which shows better performance in both robustness and fragility. In addition, multi-frequency detection is presented to provide an application-defined trade-off. To ensure the security of the proposed approach and its resistance to a chosen-plaintext attack, a randomized pixel modulation based on the Rdnyi chaotic map is employed, combining with the nonliner wave atom transform. The experimental results reveal that the proposed scheme is robust against content-preserving manipulations and has a good discriminative capability to malicions tampering.展开更多
A lexicographic image hash method based on space and frequency features was proposed.At first,the image database was constructed,and then color and texture features were extracted from the image blocks including infor...A lexicographic image hash method based on space and frequency features was proposed.At first,the image database was constructed,and then color and texture features were extracted from the image blocks including information for every image in the database,which formed feature vectors.The feature vectors were clustered to form dictionary.In hash generation,the image was preprocessed and divided into blocks firstly.Then color and texture features vectors were extracted from the blocks.These feature vectors were used to search the dictionary,and the nearest word in dictionary for each block was used to form the space features.At the same time,frequency feature was extracted from each block.The space and frequency features were connected to form the intermediate hash.Lastly,the final hash sequence was obtained by pseudo-randomly permuting the intermediate hash.Experiments show that the method has a very low probability of collision and a good perception of robustness.Compared with other methods,this method has a low collision rate.展开更多
文摘现有的深度哈希图像检索方法主要采用卷积神经网络,提取的深度特征的相似性表征能力不足.此外,三元组深度哈希主要从小批量数据中构建局部三元组样本,样本数量较少,数据分布缺失全局性,使网络训练不够充分且收敛困难.针对上述问题,文中提出基于类相似特征扩充与中心三元组损失的哈希图像检索模型(Hash Image Retrieval Based on Category Similarity Feature Expansion and Center Triplet Loss,HRFT-Net).设计基于Vision Transformer的哈希特征提取模块(Hash Feature Extraction Module Based on Vision Transformer,HViT),利用Vision Transformer提取表征能力更强的全局特征信息.为了扩充小批量训练样本的数据量,提出基于类约束的相似特征扩充模块(Similar Feature Expansion Based on Category Constraint,SFEC),利用同类样本间的相似性生成新特征,丰富三元组训练样本.为了增强三元组损失的全局性,提出基于Hadamard的中心三元组损失函数(Central Triplet Loss Function Based on Hadamard,CTLH),利用Hadamard为每个类建立全局哈希中心约束,通过增添局部约束与全局中心约束的中心三元组加速网络的学习和收敛,提高图像检索的精度.在CIFAR10、NUS-WIDE数据集上的实验表明,HRFT-Net在不同长度比特位哈希码检索上的平均精度均值较优,由此验证HRFT-Net的有效性.
文摘It is well known that robustness, fragility, and security are three important criteria of image hashing; however how to build a system that can strongly meet these three criteria is still a challenge. In this paper, a content-based image hashing scheme using wave atoms is proposed, which satisfies the above criteria. Compared with traditional transforms like wavelet transform and discrete cosine transform (DCT), wave atom transform is adopted for the sparser expansion and better characteristics of texture feature extraction which shows better performance in both robustness and fragility. In addition, multi-frequency detection is presented to provide an application-defined trade-off. To ensure the security of the proposed approach and its resistance to a chosen-plaintext attack, a randomized pixel modulation based on the Rdnyi chaotic map is employed, combining with the nonliner wave atom transform. The experimental results reveal that the proposed scheme is robust against content-preserving manipulations and has a good discriminative capability to malicions tampering.
基金Natural Science Foundations of Shanghai,China(Nos.15ZR1418500,15ZR1418400)the Training Program of Shanghai University of Electric Power for Academic Backbone Teachers,China
文摘A lexicographic image hash method based on space and frequency features was proposed.At first,the image database was constructed,and then color and texture features were extracted from the image blocks including information for every image in the database,which formed feature vectors.The feature vectors were clustered to form dictionary.In hash generation,the image was preprocessed and divided into blocks firstly.Then color and texture features vectors were extracted from the blocks.These feature vectors were used to search the dictionary,and the nearest word in dictionary for each block was used to form the space features.At the same time,frequency feature was extracted from each block.The space and frequency features were connected to form the intermediate hash.Lastly,the final hash sequence was obtained by pseudo-randomly permuting the intermediate hash.Experiments show that the method has a very low probability of collision and a good perception of robustness.Compared with other methods,this method has a low collision rate.