联合损失优化孪生网络的行人重识别

针对行人重识别应用中行人图像易受到光照、相似着装、拍摄角度影响而出现难分样本对,导致错误匹配的问题,提出一种联合损失结合孪生网络的行人重识别优化算法。首先利用残差卷积神经网络提取图像特征,并以焦点损失（Focal Loss）和交叉熵损失的联合损失对提取的特征进行监督训练,增加模型对难分样本对的关注度;然后采用余弦距离计算图像间的相似度实现行人的重识别;最后加入重排序算法降低误匹配率。采用Market-1501和DukeMTMC-reID数据集进行实验,结果表明,该算法的匹配率分别为91.2%和84.4%,平均精度均值（mAP）分别为85.8%和78.6%。     

基于局部奇异值检测的虹膜识别算法

虹膜识别以其唯一性、稳定性和非侵犯性等优点成为生物特征识别中极具发展潜力的身份识别技术;提出一种基于二次样条小波变换的虹膜识别算法;运用紧支撑的二次样条小波对构造的一维灰度信号进行分析,选择适当尺度上的小波变换结果进行量化,构造出二进制特征向量;采用Hamming距离进行移位匹配;最终完成虹膜识别;算法的错误接受率和错误拒绝率都比较低,能够达到较好的识别效果.     

基于FAR和FRR融合的多模态生物特征识别

通过多生物特征识别融合可以显著地改善系统的识别性能,在多生物特征识别中, 匹配分数级融合最常用. 现有的匹配分数级融合策略包括基于归一化的融合、基于密度的融合和基于分类器的融合. 本文分析了这三种融合策略的优缺点, 结合分数归一化和基于密度方法的优点, 提出了一种新的基于信任度的融合策略. 其中, 信任度是以错误拒绝率和错误接受率为基础, 既避免了直接求取某个匹配分数的后验概率, 又能够刻画匹配分数的分布. 将本文方法与几种有代表性的方法进行实验比较, 结果表明, 这种新融合模式可以有效地改进多生物特征识别系统的性能.     

一种特征矩阵的相似性度量方法及其在图像检索中的应用

提出一种基于DP匹配的特征矩阵相似性度量方法.首先,在对象矩阵与样本矩阵的行向量之间采用一维DP匹配方法,产生一个相似行向量来替代对象矩阵.然后再用一维DP匹配计算相似行向量与样本矩阵的标准行向量之间的匹配距离.最后在匹配距离上定义两个特征矩阵的相似度.此方法本质上是将二维特征矩阵的匹配问题转化为两个一维向量的DP匹配,适用于解决二维对象的识别和检索问题.在图像检索系统平台中对本文给出 的相似性度量方法进行验证,结果表明此方法是有效的.     

基于Gabor相位编码的手背静脉识别算法

提出了一种基于Gabor相位编码的手背静脉识别算法.该算法主要分为两部分:一个是对手背静脉图片的预处理部分,包括滤波去噪和感兴趣区域(ROI)提取;另一个是静脉特征提取和匹配识别部分,该部分主要利用静脉图像的Gabor相位编码作为静脉识别的有效特征,同时利用海明距离实现特征匹配识别.该算法应用于自制的手背静脉图库,可达到100%的识别率和0%的误识率.结果表明,该算法是一种有效的生物特征识别方法.     

基于边缘特征和像素结构相似度的图像修复算法

针对Criminisi算法的块匹配准则仅采取单一颜色判断因子导致无法合理选择最佳样本块,且其在修复过程中使用单一修复模板易出现填充裂纹和错误像素的问题,提出基于边缘特征和像素结构相似度的图像修复算法.首先提出一种局部特征与边缘纹理分辨相结合的分段修复算法以增强边缘纹理分辨能力;其次采用样本相似度和信息熵相似度确定最佳样本块集合,并依据颜色和特征项的欧氏几何距离及结构相似性确立块匹配准则;再通过基于信息熵的自适应修复模板解决Criminisi算法的填充裂纹和错误像素问题;最后引入果蝇优化算法以减少图像修复时间.实验结果证明,对于不同的图像,文中算法能取得较为满意的修复效果和修复效率.     

基于有监督双正则NMF的静脉识别算法

为使提取的静脉图像特征具有较好的聚类特性以更利于正确识别,提出了一种基于有监督非负矩阵分解的识别算法。首先,对静脉图像进行分块处理,通过融合所有的子图像特征形成静脉的原始特征;其次,采用特征的稀疏性与聚类属性双正则项,对原始的非负矩阵分解模型进行改进;然后,基于梯度下降法对改进的非负矩阵分解模型进行求解,实现对原始特征的降维与优化;最后,利用最近邻算法对新的特征进行匹配,从而获得识别结果。实验结果表明,对于3种静脉样本数据库,所提识别算法的错误接受率与错误拒绝率分别可以达到0.02与0.03;此外,其2.89s的识别时间可以满足实时性要求。     

面向法律文书的中文文本校对方法研究

在研究法律文书书写错误的语言表述特征后,将法律文书中的文本错误分为叙事陈述时的直接错误和行文书写时的隐含错误,并构建一组正则匹配规则和字词识别规则来进行错字错词识别。通过对法律文书语言学特征的研究,提出一种规则与概率统计相结合的方法实现对法律文书的文本校对。实验结果显示,该方法的召回率和准确率均达到80%,具有较好的使用前景。     

基于跳变点能量的DTW匹配在线手写签名验证算法*

提出一种以签名能量为特征的在线手写签名验证算法,侧重签名能量特征提取和匹配判决的研究.对签名波形进行小波分解,提取签名波形在跳变点处的能量,从中提取若干个能量值作为特征矢量;在基于动态时间规整的特征匹配用改进的动态时间规整方法将测试特征序列和模式特征序列进行匹配的基础上,计算最小匹配距离,得出匹配路径.实验表明, 动态时间规整算法在签名验证识别中获得了良好性能,对于随机伪造签名,误拒率为0时,误纳率为6.86%.     

基于流形学习的用户身份认证

本文基于等距映射（ISOMAP）非线性降维算法, 提出了一种新的基于用户击键特征的用户身份认证算法, 该算法用测地距离代替传统的欧氏距离, 作为样本向量之间的距离度量,在用户击键特征向量空间中挖掘嵌入的低维黎曼流形,进行用户识别。用采集到的1500个击键模式数据进行实验测试,结果表明,该文的算法性能优于现有的同类算法,其错误拒绝率（FRR）和错误通过率（FAR）分别是1.65%和0%,低于现有的同类算法。     

Ensemble of local and global information for finger-knuckle-print recognition

Biometrics authentication is an effective method for automatically recognizing a person’s identity. Recently, it has been found that the finger-knuckle-print (FKP), which refers to the inherent skin patterns of the outer surface around the phalangeal joint of one’s finger, has high capability to discriminate different individuals, making it an emerging biometric identifier. In this paper, based on the results of psychophysics and neurophysiology studies that both local and global information is crucial for the image perception, we present an effective FKP recognition scheme by extracting and assembling local and global features of FKP images. Specifically, the orientation information extracted by the Gabor filters is coded as the local feature. By increasing the scale of Gabor filters to infinite, actually we can get the Fourier transform of the image, and hence the Fourier transform coefficients of the image can be taken as the global features. Such kinds of local and global features are naturally linked via the framework of time-frequency analysis. The proposed scheme exploits both local and global information for the FKP verification, where global information is also utilized to refine the alignment of FKP images in matching. The final matching distance of two FKPs is a weighted average of local and global matching distances. The experimental results conducted on our FKP database demonstrate that the proposed local-global information combination scheme could significantly improve the recognition accuracy obtained by either local or global information and lead to promising performance of an FKP-based personal authentication system.     

基于非线性距离和夹角组合的最近特征空间嵌入方法

最近特征空间嵌入NFSE方法在训练过程中选取最近特征空间时采用传统的欧氏距离度量会导致类内离散度和类间离散度变化同步;测试时,最近邻规则也使用欧氏距离度量,而高维空间样本间直线距离具有趋同性。这些都会降低识别率,为解决此问题,提出了基于非线性距离和夹角组合的最近特征空间嵌入方法。在训练阶段,该方法使用非线性距离度量选取最近特征空间,使类内离散度的变化速度远小于类间离散度的变化速度,从而使转换空间中同类样本距离更小,不同类样本距离更大。在匹配阶段,使用结合夹角度量的最近邻分类器,充分利用样本相似性与样本夹角的关系,更适合高维空间中样本分类。仿真实验表明,基于非线性距离和夹角组合的最近特征空间嵌入方法的性能总体上优于对比算法。     

单应性矩阵自适应估计方法

如何从初始匹配点集中估计出精确的单应性矩阵,有效地剔除误匹配,一直以来都是视觉领域研究的重点和难点,也是实际相关技术应用中最为关键的一步。通过将特征点对相似度概念应用于LMedS的样本选取过程,提出了一种新的单应性矩阵自适应的估计方法。区别于传统LMeds方法从无序匹配点集中随机选取样本的过程,该方法首先以点对间的相似度对整个初始匹配点进行降序排列,然后从前往后依次选取样本。实验结果表明,与LMedS相比,该方法估计出的单应性矩阵更精确、鲁棒,效率更高（得到最佳模型所需的迭代次数仅约为LMedS的1/5）,同时弥补了RANSAC及其改进方法需预先设置距离偏差阈值的不足。     

Unsupervised manifold learning based on multiple feature spaces

Manifold learning is a well-known dimensionality reduction scheme which can detect intrinsic low-dimensional structures in non-linear high-dimensional data. It has been recently widely employed in data analysis, pattern recognition, and machine learning applications. Isomap is one of the most promising manifold learning algorithms, which extends metric multi-dimensional scaling by using approximate geodesic distance. However, when Isomap is conducted on real-world applications, it may have some difficulties in dealing with noisy data. Although many applications represent a special sample by multiple feature vectors in different spaces, Isomap employs samples in unique observation space. In this paper, two extended versions of Isomap to multiple feature spaces problem, namely fusion of dissimilarities and fusion of geodesic distances, are presented. We have employed the advantages of several spaces and depicted the Euclidean distance on learned manifold that is more compatible to the semantic distance. To show the effectiveness and validity of the proposed method, some experiments have been carried out on the application of shape analysis on MPEG7 CE Part B and Fish data sets.     

Online finger-knuckle-print verification for personal authentication

Biometric based personal authentication is an effective method for automatically recognizing, with a high confidence, a person's identity. By observing that the texture pattern produced by bending the finger knuckle is highly distinctive, in this paper we present a new biometric authentication system using finger-knuckle-print (FKP) imaging. A specific data acquisition device is constructed to capture the FKP images, and then an efficient FKP recognition algorithm is presented to process the acquired data in real time. The local convex direction map of the FKP image is extracted based on which a local coordinate system is established to align the images and a region of interest is cropped for feature extraction. For matching two FKPs, a feature extraction scheme, which combines orientation and magnitude information extracted by Gabor filtering is proposed. An FKP database, which consists of 7920 images from 660 different fingers, is established to verify the efficacy of the proposed system and promising results are obtained. Compared with the other existing finger-back surface based biometric systems, the proposed FKP system achieves much higher recognition rate and it works in real time. It provides a practical solution to finger-back surface based biometric systems and has great potentials for commercial applications.     

Encoding local image patterns using Riesz transforms: With applications to palmprint and finger-knuckle-print recognition

Biometrics authentication is an effective method for automatically recognizing a person's identity with high confidence. It is well recognized that in biometric systems feature extraction and representation are key considerations. Among various feature extraction and representation schemes, coding-based methods are most attractive because they have the merits of high accuracy, robustness, compactness and high matching speed, and thus they have been adopted in many different kinds of biometric systems, such as iris, palmprint, and finger-knuckle-print based ones. However, how to devise a good coding scheme is still an open issue. Recent studies in image processing and applied mathematics have shown that local image features can be well extracted with Riesz transforms in a unified framework. Thus, in this paper we propose to utilize Riesz transforms to encode the local patterns of biometric images. Specifically, two Riesz transforms based coding schemes, namely RCode1 and RCode2, are proposed. They both use 3-bits to represent each code and employ the normalized Hamming distance for matching. RCode1 and RCode2 are thoroughly evaluated and compared with the other 3-bit coding methods on a palmprint database and a finger-knuckle-print database. Experiments show that the proposed methods, especially RCode2, could achieve quite similar verification accuracies with the state-of-the-art method (CompCode) while they need much less time at the feature extraction stage, which renders them better candidates for time critical applications.     

高光谱目标表述的不均衡数据多示例学习方法

高光谱目标表述是高光谱目标检测中的核心问题。在众多高光谱目标表述方法中, 多示例学习方法(MIL)由于不需要精确的像素级语义标签等因素,而成为研究高光谱目标表述的 一个有效方法。但是,面向高光谱目标表述的多示例学习方法中,存在正包内目标示例远少于 背景示例的示例级数据不均衡问题,导致学习到的目标表述性能不佳。为此,提出一种面向不 均衡数据的多示例学习方法,提取每个包中最可能为正的示例组成正示例集,以此为基础合成 新的正样本,增加正样本在正包中所占比例,改善高光谱目标表述能力。在真实高光谱数据上 验证所提方法的有效性,结果表明该方法使正包样本组成更均衡,从而学习到更正确的目标表 述,提高目标检测的性能。     

面向移动社会网络的好友推荐方法

针对移动社会网络（MSN）的好友推荐问题,提出了一种基于多维相似度的好友推荐方法。该方法隶属于基于内容的好友推荐,但与现有方法相比,不再局限于单一维度的匹配信息,而是从空间、时间和兴趣三个维度出发,判断用户在各个维度上的相似度,最终通过“差异距离”进行综合评判,向目标用户推荐与之在地理位置、在线时间和兴趣爱好上更具一致性的其他用户成为其好友。由实验结果表明,该方法应用于移动社会网络中的好友推荐服务时,其推荐结果查准率接近80%,查准效率接近60%,性能远高于只基于单一维度的好友推荐方法;同时,通过对三维权重值的调整,该方法可应用于多种特性的移动社会网络中。     

Phase congruency induced local features for finger-knuckle-print recognition

Researchers have recently found that the finger-knuckle-print (FKP), which refers to the inherent skin patterns of the outer surface around the phalangeal joint of one’s finger, has high discriminability, making it an emerging promising biometric identifier. Effective feature extraction and matching plays a key role in such an FKP based personal authentication system. This paper studies image local features induced by the phase congruency model, which is supported by strong psychophysical and neurophysiological evidences, for FKP recognition. In the computation of phase congruency, the local orientation and the local phase can also be defined and extracted from a local image patch. These three local features are independent of each other and reflect different aspects of the image local information. We compute efficiently the three local features under the computation framework of phase congruency using a set of quadrature pair filters. We then propose to integrate these three local features by score-level fusion to improve the FKP recognition accuracy. Such kinds of local features can also be naturally combined with Fourier transform coefficients, which are global features. Experiments are performed on the PolyU FKP database to validate the proposed FKP recognition scheme.     

基于特征点动态选择的三维人脸点云模型重建

针对典型的点云配准方法中伪特征点过多导致配准效率低和配准结果不精确的问题,提出一种基于特征点动态选择的三维人脸点云模型重建方法。该方法在粗配准阶段,采用动态特征矩阵求解法获取粗匹配特征变换矩阵以避免伪特征点的干扰。在精配准过程中,采用二次加权法向量垂直距离法在人脸流形表面选择更有效的特征点以减少伪特征点的数量,并采用基于特征融合与局部特征一致性的迭代最近点方法进行精配准。经过对比实验验证了算法的可行性,实验结果表明,提出算法能够实现高精度且快速的三维人脸点云模型重建,且均方根误差达到1.816 5 mm,相较于其他算法,在模型重建精度和效率方面都有所提升,具有良好的应用前景。