An efficient dynamic reliability-dependent bit allocation for biometric discretization

A biometric discretization scheme converts biometric features into a binary string via segmenting every one-dimensional feature space into multiple labelled intervals, assigning each interval-captured feature element with a short binary string and concatenating the binary output of all feature elements into a bit string. This paper proposes a bit allocation algorithm for biometric discretization to allocate bits dynamically to every feature element based on a Binary Reflected Gray code. Unlike existing bit allocation schemes, our scheme bases upon a combination of bit statistics (reliability measure) and signal to noise ratio (discriminability measure) in performing feature selection and bit allocation procedures. Several empirical comparative studies are conducted extensively on two popular face datasets to justify the efficiency and feasibility of our proposed approach.     

Face and palmprint pixel level fusion and Kernel DCV-RBF classifier for small sample biometric recognition

Recently, multi-modal biometric fusion techniques have attracted increasing atove the recognition performance in some difficult biometric problems. The small sample biometric recognition problem is such a research difficulty in real-world applications. So far, most research work on fusion techniques has been done at the highest fusion level, i.e. the decision level. In this paper, we propose a novel fusion approach at the lowest level, i.e. the image pixel level. We first combine two kinds of biometrics: the face feature, which is a representative of contactless biometric, and the palmprint feature, which is a typical contacting biometric. We perform the Gabor transform on face and palmprint images and combine them at the pixel level. The correlation analysis shows that there is very small correlation between their normalized Gabor-transformed images. This paper also presents a novel classifier, KDCV-RBF, to classify the fused biometric images. It extracts the image discriminative features using a Kernel discriminative common vectors (KDCV) approach and classifies the features by using the radial base function (RBF) network. As the test data, we take two largest public face databases (AR and FERET) and a large palmprint database. The experimental results demonstrate that the proposed biometric fusion recognition approach is a rather effective solution for the small sample recognition problem.     

Robust memory-efficient data level information fusion of multi-modal biometric images

This paper presents a novel multi-level wavelet based fusion algorithm that combines information from fingerprint, face, iris, and signature images of an individual into a single composite image. The proposed approach reduces the memory size, increases the recognition accuracy using multi-modal biometric features, and withstands common attacks such as smoothing, cropping, JPEG 2000, and filtering due to tampering. The fusion algorithm is validated using the verification algorithms we developed, existing algorithms, and commercial algorithm. In addition to our multi-modal database, experiments are also performed on other well known databases such as FERET face database and CASIA iris database. The effectiveness of the fusion algorithm is experimentally validated by computing the matching scores and the equal error rates before fusion, after reconstruction of biometric images, and when the composite fused image is subjected to both frequency and geometric attacks. The results show that the fusion process reduced the memory required for storing the multi-modal images by 75%. The integrity of biometric features and the recognition performance of the resulting composite fused image is not affected significantly. The complexity of the fusion and the reconstruction algorithms is O(n log n) and is suitable for many real-time applications. We also propose a multi-modal biometric algorithm that further reduces the equal error rate compared to individual biometric images.     

模糊逻辑在人脸特征保护算法中的应用

随着人脸识别在门禁、视频监控等公共安全领域中的应用日益广泛,人脸特征数据的安全性和隐私性问题成为备受关注的焦点。近年来出现了许多关于生物特征及人脸特征的安全保护算法,这些算法大都是将生物特征数据转变为二值的串,再进行保护。针对已有的保护算法中将实值的人脸特征转换为二值的串,从而导致信息丢失的不足,应用模糊逻辑对人脸模板数据的类内差异进行建模,从而提高人脸识别系统的性能。给出了算法在CMU PIE的光照子集、CMU PIE带光照和姿势的子集和ORL人脸数据库中的实验结果。实验表明,该算法能够进一步提高已有安全保护算法的识别率。     

基于多特征模糊聚类的图像融合方法

首先利用模糊C-均值聚类算法在多特征形成的特征空间上对图像进行区域分割,并在此基础上对区域进行多尺度小波分解;然后利用柯西函数构造区域的模糊相似度,应用模糊相似度及区域信息量构造加权因子,从而得到融合图像的小波系数;最后利用小波逆变换得到融合图像.采用均方根误差、峰值信噪比、熵、交叉熵和互信息5种准则评价融合算法的性能.实验结果表明,文中方法具有良好的融合特性.     

Confidence factor weighted Gaussian function induced parallel fuzzy rank-level fusion for inference and its application to face recognition

This paper proposes a novel approach for inference using fuzzy rank-level fusion and explores it application to face recognition using multiple biometric representations. Multiple representations of single biometric (trait) aim to increase the reliability or acceptance of a biometric system, as it exploits the underlying essential characteristics provided by different sensors. In this paper, we propose a new scheme for generating fuzzy ranks induced by a Gaussian function based on the confidence of a classifier. In contrast to the conventional ranking, this fuzzy ranking reflects some associations among the outputs (confidence factors) of a classifier. These fuzzy ranks, yielded by multiple representations of a face image, are fused weighted by the corresponding confidence factors of the classifier to generate the final ranks while recognizing a face. In many real-world applications, where multiple traits of a person are unavailable, the proposed method is highly effective. However, it can easily be extended to multimodal biometric systems utilizing multiple classifiers. The experimental results using different feature vectors of a face image employing different classifiers show that the proposed method can significantly improve recognition accuracy as compared to those from individual feature vectors and as well as some commonly used rank-level fusion methods.     

预训练模型特征提取的双对抗磁共振图像融合网络研究

随着多模态医学图像在临床诊疗工作中的普及,建立在时空相关性特性基础上的融合技术得到快速发展,融合后的医学图像不仅可以保留各模态源图像的独有特征,而且能够强化互补信息、便于医生阅片.目前大多数方法采用人工定义约束的策略来实现特征提取和特征融合,这容易导致融合图像中部分有用信息丢失和细节不清晰等问题.为此,提出一种基于预训练模型特征提取的双对抗融合网络实现MR-T1/MR-T2图像的融合.该网络由一个特征提取模块、一个特征融合模块和两个鉴别网络模块组成.由于已配准的多模态医学图像数据集规模较小,无法对特征提取网络进行充分的训练,又因预训练模型具有强大的数据表征能力,故将预先训练的卷积神经网络模型嵌入到特征提取模块以生成特征图.然后,特征融合网络负责融合深度特征并输出融合图像.两个鉴别网络通过对源图像与融合图像进行准确分类,分别与特征融合网络建立对抗关系,最终激励其学习出最优的融合参数.实验结果证明了预训练技术在所提方法中的有效性,同时与现有的6种典型融合方法相比,所提方法融合结果在视觉效果和量化指标方面均取得最优表现.     

基于行为足迹的多模态融合身份认证

针对单模态身份认证方法存在特征单一容易被伪造和攻破的问题,提出基于用户行为足迹的多模态特征融合隐式身份认证方法。在移动设备中采集用户使用设备时的触摸压力、触摸轨迹、加速度等传感器数据,利用特征选择技术提取触摸屏交互、移动模式、物理位置等特征并对其进行训练与融合,最终通过多模态特征融合模型实现用户身份认证。实验结果表明,该方法采用的特征级融合和决策级融合方式均获得了98%以上的认证准确率,相比单模态身份认证方法更难以被伪造和攻破,且认证准确率更高、稳定性更强。     

熵权法融合局部匹配Gabor特征的鲁棒人脸识别

针对复杂环境下人脸识别难度大的问题,提出了一种熵权法融合局部Gabor特征方法。计算类熵加权向量;计算局部归一化输入图像的Borda计数矩阵,从而消除低值Gabor jet比较矩阵;通过将分数层类熵加权Gabor特征与LGBP和LGXP融合解决了完成人脸的识别。在FERET、AR和FRGC 2.0人脸数据库上的实验结果表明,该方法对轻微姿态变化具有显著鲁棒性,并且对人眼检测中高达3像素的误差具有鲁棒性,相比其他几种人脸识别方法,该方法取得了更好的识别效果。     

基于深度学习的虹膜人脸多特征融合识别

传统多生物特征融合识别方法中人工设计特征提取存在盲目性和差异性,特征融合存在空间不匹配或维度过高等问题,为此提出一种基于深度学习的多生物特征融合识别方法。通过卷积神经网络(convolutional neural networks,CNN)提取人脸和虹膜特征、参数化t-SNE算法特征降维和支持向量机(support vector machine,SVM)分类组合进行融合识别。实验结果表明,该融合识别方法与单一生物特征识别以及其它融合识别方法相比,鲁棒性增强,识别性能提升明显。     

An effective biometric cryptosystem combining fingerprints with error correction codes

With the emergence and popularity of identity verification means by biometrics, the biometric system which can assure security and privacy has received more and more concentration from both the research and industry communities. In the field of secure biometric authentication, one branch is to combine the biometrics and cryptography. Among all the solutions in this branch, fuzzy commitment scheme is a pioneer and effective security primitive. In this paper, we propose a novel binary length-fixed feature generation method of fingerprint. The alignment procedure, which is thought as a difficult task in the encrypted domain, is avoided in the proposed method due to the employment of minutiae triplets. Using the generated binary feature as input and based on fuzzy commitment scheme, we construct the biometric cryptosystems by combining various of error correction codes, including BCH code, a concatenated code of BCH code and Reed-Solomon code, and LDPC code. Experiments conducted on three fingerprint databases, including one in-house and two public domain, demonstrate that the proposed binary feature generation method is effective and promising, and the biometric cryptosystem constructed by the feature outperforms most of the existing biometric cryptosystems in terms of ZeroFAR and security strength. For instance, in the whole FVC2002 DB2, a 4.58% ZeroFAR is achieved by the proposed biometric cryptosystem with the security strength 48 bits.     

基于NSCT与模糊逻辑的图像融合方法

提出一种基于非下采样Contourlet变换（NSCT）和模糊逻辑的图像融合方法。NSCT分解有利于更好地保持图像的边缘信息和轮廓结构,并增强图像的平移不变性。在对原图像进行多尺度几何分解后,针对图像融合过程中原图像融入信息程度的不确定性,采用基于模糊逻辑（Fuzzy logic）的融合规则指导图像的融合：根据高低频各自的频带特性和待融合图像的特点,对高频使用基于信息熵与模糊逻辑的融合方法,而对低频使用基于亮度、梯度、方差、信息熵等特性及模糊逻辑的规则得到融合系数,最终经过NSCT逆变换得到融合图像。实验结果表明,该算法的融合图像具有良好的视觉效果及客观评指标,指标与视觉效果也能够较好的统一。     

基于矩阵分解双线性池化的多模态融合虚假新闻检测

现有的大多数虚假新闻检测方法将视觉和文本特征串联拼接,导致模态信息冗余并且忽略了不同模态信息之间的相关性。为了解决上述问题,提出一种基于矩阵分解双线性池化的多模态融合虚假新闻检测算法。首先,该算法将多模态特征提取器捕捉的文本和视觉特征利用矩阵分解双线性池化方法进行有效融合,然后与虚假新闻检测器合作鉴别虚假新闻;此外,在训练阶段加入了事件分类器来预测事件标签并去除事件相关的依赖。在Twitter和微博两个多模态谣言数据集上进行了对比实验,证明了该算法的有效性。实验结果表明提出的模型能够有效地融合多模态数据,缩小模态间的异质性差异,从而提高虚假新闻检测的准确性。     

基于同一特征空间的多模态脑肿瘤分割方法

脑胶质瘤的分割依赖多种模态的核磁共振成像（MRI）的影像。基于卷积神经网络（CNN）的分割算法往往是在固定的多种模态影像上进行训练和测试,这忽略了模态数据缺失或增加问题。针对这个问题,提出了将不同模态的图像通过CNN映射到同一特征空间下并利用同一特征空间下的特征来分割肿瘤的方法。首先,不同模态的数据经过同一深度CNN提取特征;然后,将不同模态的特征连接起来,经过全连接层实现特征融合;最后,利用融合的特征实现脑肿瘤分割。模型采用BRATS2015数据集进行训练和测试,并使用Dice系数对模型进行验证。实验结果表明了所提模型能有效缓解数据缺失问题。同时,该模型较多模态联合的方法更加灵活,能够应对模态数据增加问题。     

Memory based fusion for multi-modal deep learning

The use of multi-modal data for deep machine learning has shown promise when compared to uni-modal approaches with fusion of multi-modal features resulting in improved performance in several applications. However, most state-of-the-art methods use naive fusion which processes feature streams independently, ignoring possible long-term dependencies within the data during fusion. In this paper, we present a novel Memory based Attentive Fusion layer, which fuses modes by incorporating both the current features and long-term dependencies in the data, thus allowing the model to understand the relative importance of modes over time. We introduce an explicit memory block within the fusion layer which stores features containing long-term dependencies of the fused data. The feature inputs from uni-modal encoders are fused through attentive composition and transformation followed by naive fusion of the resultant memory derived features with layer inputs. Following state-of-the-art methods, we have evaluated the performance and the generalizability of the proposed fusion approach on two different datasets with different modalities. In our experiments, we replace the naive fusion layer in benchmark networks with our proposed layer to enable a fair comparison. Experimental results indicate that the MBAF layer can generalize across different modalities and networks to enhance fusion and improve performance.     

基于全局特征改进的行人重识别

由于行人重识别面临姿态变化、遮挡干扰、光照差异等挑战, 因此提取判别力强的行人特征至关重要. 本文提出一种在全局特征基础上进行改进的行人重识别方法, 首先, 设计多重感受野融合模块充分获取行人上下文信息, 提升全局特征辨别力; 其次, 采用GeM池化获取细粒度特征; 最后, 构建多分支网络, 融合网络不同深度的特征预测行人身份. 本文方法在Market1501和DukeMTMC-ReID两大数据集上的mAP指标分别达到83.8%和74.9%. 实验结果表明, 本文方法有效改进了基于全局特征的模型, 提升了行人重识别的识别准确率.     

A secure biometric discretization scheme for face template protection

In this paper, a dynamic biometric discretization scheme based on Linnartz and Tuyls’s quantization index modulation scheme (LT-QIM) [Linnartz and Tuyls, 2003] is proposed. LT-QIM extracts one bit per feature element and takes care of the intra-class variation of the biometric features. Nevertheless, LT-QIM does not consider statistical distinctiveness between users, and thus lacks the capability of preserving the discriminative power of the original biometric features. We put forward a generalized LT-QIM scheme in such a way that it allocates multiple bits to each feature element according to a statistical distinctiveness measure of the feature. Hence, more bits are assigned to high distinctive features and fewer bits to low distinctive features. With provision for intra-class variation compensation and dynamic bit allocation by means of the statistical distinctiveness measure, the generalized scheme enhances the verification performance compared to the original scheme. Several comparative studies are conducted on two popular face data sets to justify the efficiency and feasibility of our proposed scheme. The security aspect is also considered by including a stolen-token scenario.     

基于非向下采样contourlet的自适应图像融合算法

针对融合后图像模糊现象,提出一种基于非向下采样contourlet的自适应图像融合算法.分析了轮廓波变换和非抽样轮廓波变换的原理,采用非向下采样contourlet对图像进行分解,依据低频变化设置阈值来调节低频变化率和均匀度在决策规则中所占的比例.当低频变化率之差高于阈值时,采用基于均匀度的融合规则;当低频变化率之差低于阈值时,采用基于变化率的融合规则.对于高频部分则采用高频系数对比度的处理策略.通过熵、相对误差和清晰度对实验结果进行了评价,结果表明,基于非向下采样contourlet的自适应融合算法取得了良好的融合效果.     

自适应多模态特征融合胶质瘤分级网络

目的 胶质瘤的准确分级是辅助制定个性化治疗方案的主要手段,但现有研究大多数集中在基于肿瘤区域的分级预测上,需要事先勾画感兴趣区域,无法满足临床智能辅助诊断的实时性需求。因此,本文提出一种自适应多模态特征融合网络（adaptive multi-modal fusion net,AMMFNet）,在不需要勾画肿瘤区域的情况下,实现原始采集图像到胶质瘤级别的端到端准确预测。方法 AMMFNet方法采用4个同构异义网络分支提取不同模态的多尺度图像特征;利用自适应多模态特征融合模块和降维模块进行特征融合;结合交叉熵分类损失和特征嵌入损失提高胶质瘤的分类精度。为了验证模型性能,本文采用MICCAI （Medical Image Computing and Computer Assisted Intervention Society）2018公开数据集进行训练和测试,与前沿深度学习模型和最新的胶质瘤分类模型进行对比,并采用精度以及受试者曲线下面积（area under curve,AUC）等指标进行定量分析。结果 在无需勾画肿瘤区域的情况下,本文模型预测胶质瘤分级的AUC为0.965;在使用肿瘤区域时,其AUC高达0.997,精度为0.982,比目前最好的胶质瘤分类模型——多任务卷积神经网络同比提高1.2%。结论 本文提出的自适应多模态特征融合网络,通过结合多模态、多语义级别特征,可以在未勾画肿瘤区域的前提下,准确地实现胶质瘤分级预测。     

Optimal sequential fusion for multibiometric cryptosystems

Biometric cryptosystems have been widely studied in the literature to protect biometric templates. To ensure sufficient security of the biometric cryptosystem against the offline brute-force attack (also called the FAR attack), it is critical to reduce FAR of the system. One of the most effective approaches to improve the accuracy is multibiometric fusion, which can be divided into three categories: feature level fusion, score level fusion, and decision level fusion. Among them, only feature level fusion can be applied to the biometric cryptosystem for security and accuracy reasons. Conventional feature level fusion schemes, however, require a user to input all of the enrolled biometric samples at each time of authentication, and make the system inconvenient.In this paper, we first propose a general framework for feature level sequential fusion, which combines biometric features and makes a decision each time a user inputs a biometric sample. We then propose a feature level sequential fusion algorithm that can minimize the average number of input, and prove its optimality theoretically. We apply the proposed scheme to the fuzzy commitment scheme, and demonstrate its effectiveness through experiments using the finger-vein dataset that contains six fingers from 505 subjects. We also analyze the security of the proposed scheme against various attacks: attacks that exploit the relationship between multiple protected templates, the soft-decoding attack, the statistical attack, and the decodability attack.