Fine-grained cooperative access control scheme with hidden policies

The traditional ciphertext policy attribute-based encryption (CP-ABE) has two problems:one is that the access policy must be embedded in the ciphertext and sent, which leads to the disclosure of user爷 s privacy information, the other is that it does not support collaborative decryption, which cannot meet the actual demand of conditional collaborative decryption among multiple users. In order to deal with the above two problems at the same time, a fine-grained cooperative access control scheme with hidden policies (FCAC-HP) is proposed based on the existing CP-ABE schemes combined with blockchain technology. In FCAC-HP scheme, users are grouped by group identifier so that only users within the same group can cooperate. In the data encryption stage, the access policy is encrypted and then embedded in the ciphertext to protect the privacy information of the access policy. In the data access stage, the anonymous attribute matching technology is introduced so that only matched users can decrypt ciphertext data to improve the efficiency of the system. In this process, a smart contract is used to execute the verification algorithm to ensure the credibility of the results. In terms of security, FCAC-HP scheme is based on the prime subgroup discriminative assumption and is proved to be indistinguishable under chosen plaintext attack (CPA) by dual system encryption technology. Experimental verification and analysis show that FCAC-HP scheme improves computational efficiency while implementing complex functions.     

Sensitive attribute privacy preservation of trajectory data publishing based on l-diversity

Distributed and Parallel Databases - The widely application of positioning technology has made collecting the movement of people feasible for knowledge-based decision. Data in its original form...     

强隐私保护的移动群智感知方案

为了实现移动群智感知系统中身份隐私、证书撤销和积分激励功能,同时解决恶意用户身份追踪与诚实用户隐私保护之间的矛盾,提出了强隐私保护的移动群智感知方案.基于门限密码思想将身份追踪能力分散到多个实体上,使得多个追踪者合作才能追踪用户身份;将Pointcheval-Sanders签名和基于RSA假设的Camenisch-Lysyanskaya累加器结合起来实现了证书的安全快速撤销;利用Pointcheval-Sanders签名构造了保护隐私的积分激励机制.对该方案进行了安全性分析和实验分析.研究结果表明,该方案不仅满足安全要求,而且在实际应用中具有可行性.     

A lossless secret image sharing scheme using a larger finite field

Recent research has made an effort to take 8b-bit value as a polynomial coefficient and use a random number as the maximum coefficient term in a Shamir’s polynomial, where b > 0. These can help improve computationl efficiency by reducing the sum of calculating polynomials, and avoid the case of the coefficient of x^k??1 being zero. However, such research still has the issues of requiring much extra storage space, lossy secret image, shadow images with large size, and storing permutation key. To solve the above issues, in this paper, we propose a novel scheme which takes 8b-bit value as a polynomial coefficient, designs a bit-level method and runs under Galois Field GF(2^8b). Experimental results show that this scheme improves existing similar schemes on several aspects, such as less extra storage space and higher computational performance.

     

面向非平衡数据集的金融欺诈账户检测研究

针对非平衡金融数据集,提出一种银行欺诈账户检测框架iForest-SMOTE。基于账户的动态交易特点,从统计、时序、监督信息维度抽取账户交易行为特征。针对过采样技术ADASYN在金融账户数据集中存在的跨区域样本合成问题,提出一种基于iForest算法的数据集均衡预处理策略,通过iForest算法对数据进行混合采样,在去除多数类噪声数据的同时降低分类器对少数类的学习难度。在此基础上,设计随机森林分类器实现金融欺诈账户检测。在真实金融账户交易数据集上进行实验,结果表明,与ADASYN、SMOTE等采样技术相比,iForest-SMOTE在召回率和准确率方面具有明显优势,F-value值至少能够提升2.13个百分点。     

基于改进的双向长短期记忆网络的视频摘要生成模型

针对传统视频摘要方法往往没有考虑时序信息以及提取的视频特征过于复杂、易出现过拟合现象的问题,提出一种基于改进的双向长短期记忆（BiLSTM）网络的视频摘要生成模型。首先,通过卷积神经网络（CNN）提取视频帧的深度特征,而且为了使生成的视频摘要更具多样性,采用BiLSTM网络将深度特征识别任务转换为视频帧的时序特征标注任务,让模型获得更多上下文信息;其次,考虑到生成的视频摘要应当具有代表性,因此通过融合最大池化在降低特征维度的同时突出关键信息以淡化冗余信息,使模型能够学习具有代表性的特征,而特征维度的降低也减少了全连接层需要的参数,避免了过拟合问题;最后,预测视频帧的重要性分数并转换为镜头分数,以此选取关键镜头生成视频摘要。实验结果表明,在标准数据集TvSum和SumMe上,改进后的视频摘要生成模型能提升生成视频摘要的准确性;而且它的F1-score值也比基于长短期记忆（LSTM）网络的视频摘要模型DPPLSTM在两个数据集上分别提高1.4和0.3个百分点。     

基于自动编码器的深度伪造图像检测方法

基于深度学习的图像伪造方法生成的图像肉眼难辨,一旦该技术被滥用于制作虚假图像和视频,可能会对国家政治、经济、文化造成严重的负面影响,也可能会对社会生活和个人隐私构成威胁。针对上述问题,提出了一种基于自动编码器的深度伪造Deepfake图像检测方法。首先,借助高斯滤波对图像进行预处理,提取高频信息作为模型输入;然后,利用自动编码器对图像进行特征提取,并在编码器中添加注意力机制模块以获取更好的分类效果;最后,通过消融实验证明,采用所提的预处理方法和添加注意力机制模块有助于伪造图像检测。实验结果表明,与ResNet50、Xception以及InceptionV3相比,所提方法在数据集样本量较小且包含的场景丰富时,可以有效检测多种生成方法所伪造的图像,其平均准确率可达97.10%,明显优于对比方法,且其泛化性能也明显优于对比方法。     

深度学习场景文本检测方法综述

自然场景文本检测对于机器理解场景等有着重要作用。近年来,随着深度学习的发展,自然场景文字检测方法也日新月异,取得了很好的检测效果。分析、总结了近年来基于深度学习的场景文字检测方法,将其归纳分类为基于回归、基于分割,以及两者混合三种类型,并对各类检测方法的优缺点进行了对比分析。介绍了场景文本检测性能指标及常用的公开数据集以及下载方式。对场景文字检测领域研究进行总结和展望,有望为深度学习场景文本检测方法提供新的研究方向。     

属性基加密和区块链结合的可信数据访问控制方案

传统的数据存储方式往往采用集中式架构,这种集中式存储架构容易产生信任和安全问题。文章提出一种属性基加密和区块链结合的可信数据访问控制方案,将对称加密算法和属性基加密算法结合,实现了数据所有者对数据的细粒度访问控制,保障了数据所有者的隐私保护权利。同时,文章将区块链技术和分布式存储相结合,区块链上仅存储数据及密文位置的哈希值和访问控制策略等信息,解决了区块链系统的存储瓶颈。实验表明,该方案能有效保护数据所有者的安全和隐私,实现数据的安全共享。     

Task-wise attention guided part complementary learning for few-shot image classification

Science China Information Sciences -