Feature subset selection applied to model-free gait recognition

In this paper, we tackle the problem of gait recognition based on the model-free approach. Numerous methods exist; they all lead to high dimensional feature spaces. To address the problem of high dimensional feature space, we propose the use of the Random Forest algorithm to rank features' importance. In order to efficiently search throughout subspaces, we apply a backward feature elimination search strategy. Our first experiments are carried out on unknown covariate conditions. Our first results suggest that the selected features contribute to increase the CCR of different existing classification methods. Secondary experiments are performed on unknown covariate conditions and viewpoints. Inspired by the location of our first experiments' features, we proposed a simple mask. Experimental results demonstrate that the proposed mask gives satisfactory results for all angles of the probe and consequently is not view specific. We also show that our mask performs well when an uncooperative experimental setup is considered as compared to the state-of-the art methods. As a consequence, we propose a panoramic gait recognition framework on unknown covariate conditions. Our results suggest that panoramic gait recognition can be performed under unknown covariate conditions. Our approach can greatly reduce the complexity of the classification problem while achieving fair correct classification rates when gait is captured with unknown conditions.     

面向开集识别的稳健测试时适应方法

开集识别旨在研究测试阶段突现未见类别对于机器学习模型的挑战,以期学习模型既能分类已见类别又可识别/拒绝未见类别,是确保机器学习模型能够在开放世界中高效稳健部署的重要技术.既有开集识别技术通常假设已见类别的协变量分布在训练与测试阶段维持不变.然而在实际场景中,类别的协变量分布常不断变化.直接利用既有技术不再奏效,其性能甚至劣于基线方案.因此,亟需研究新型开集识别方法,使其能不断适应协变量分布偏移,以期模型在测试阶段既能稳健分类已见类别又可识别未见类别.我们将此新问题设置命名为开放世界适应问题(简称AOW)并提出了一种开放测试时适应方法(简称OTA).本文方法基于无标注测试数据优化自适应熵损失与开集熵损失更新模型,维持对已见类的既有判别能力,同时增强了识别未见类的能力.大量实验分析表明,本文方法在多组基准数据集、多组不同协变量偏移程度下均稳健地优于现有先进的开集识别方法.     

On solving the face recognition problem with one training sample per subject

The lack of adequate training samples and the considerable variations observed in the available image collections due to aging, illumination and pose variations are the two key technical barriers that appearance-based face recognition solutions have to overcome. It is a well-documented fact that their performance deteriorates rapidly when the number of training samples is smaller than the dimensionality of the image space. This is especially true for face recognition applications where only one training sample per subject is available. In this paper, a recognition framework based on the concept of the so-called generic learning is introduced as an attempt to boost the performance of traditional appearance-based recognition solutions in the one training sample application scenario. Different from contemporary approaches, the proposed solution learns the intrinsic properties of the subjects to be recognized using a generic training database which consists of images from subjects other than those under consideration. Many state-of-the-art face recognition solutions can be readily integrated in the proposed framework. A novel multi-learner framework is also proposed to further boost recognition performance.Extensive experimentation reported in the paper suggests that the proposed framework provides a comprehensive solution and achieves lower error recognition rate when considered in the context of one training sample face recognition problem.     

基于自编码器和稀疏表示的单样本人脸识别

单样本人脸识别因其在现实生活中的广泛应用而成为人脸识别领域的热门话题。单张训练样本条件下训练样本的缺少和复杂的类内人脸表情、光照、遮挡变化给单样本人脸识别研究带来困难。传统的基于稀疏表示的人脸识别方法需要大量的训练样本构成过完备的字典,因而在单样本条件下识别效果明显下滑。针对这一问题,提出一种基于有监督自编码器的带变化人脸样本生成方法,在保留身份信息的同时自动生成带变化的人脸图像用于单样本条件下的字典扩充,一定程度上缓解了单样本条件下的欠采样问题,弥补了训练集和测试集间的人脸变化信息差异,使得传统的稀疏表示方法能够适用于单样本人脸识别问题。在公共数据库上的实验结果不仅证明了该方法的有效性,而且对测试集中不同的人脸变化也展现出了较强的鲁棒性。     

Double-cohesion learning based multiview and discriminant palmprint recognition

Palmprint recognition has been widely used in security authentication. However, most of the existing palmprint representation methods are focused on a special application scenario using the hand-crafted features from a single-view. If the features become weak as the application scenario changes, the recognition performance will be degraded. To address this problem, we propose to comprehensively exploit palmprint features from multiple views to improve the recognition performance in generic scenarios. In this paper, a novel double-cohesion learning based multiview and discriminant palmprint recognition (DC_MDPR) method is proposed, which imposes a double-cohesion strategy to reduce the inter-view margins for each subject and the intra-class margins for each view. In this way, for each subject, the features from different views can be closer to each other in the binary-label space. Meanwhile, for each view, the features sharing the same label information can move towards each other by imposing a neighbor graph regularization. The proposed method can be flexibly applied to any type of palmprint feature fusion. Moreover, it presents the multiview features in a low-dimensionality sub-space, effectively reducing the computational complexity. Experimental results on various palmprint databases have shown that the proposed method can always achieve the best recognition performance compared to other state-of-the-art algorithms.     

基于小样本学习的SAR图像识别

深度学习已成为图像识别领域的一个研究热点。与传统图像识别方法不同,深度学习从大量数据中自动学习特征,并且具有强大的自学习能力和高效的特征表达能力。但在小样本条件下,传统的深度学习方法如卷积神经网络难以学习到有效的特征,造成图像识别的准确率较低。因此,提出一种新的小样本条件下的图像识别算法用于解决SAR图像的分类识别。该算法以卷积神经网络为基础,结合自编码器,形成深度卷积自编码网络结构。首先对图像进行预处理,使用2D Gabor滤波增强图像,在此基础上对模型进行训练,最后构建图像分类模型。该算法设计的网络结构能自动学习并提取小样本图像中的有效特征,进而提高识别准确率。在MSTAR数据集的10类目标分类中,选择训练集数据中10%的样本作为新的训练数据,其余数据为验证数据,并且,测试数据在卷积神经网络中的识别准确率为76.38%,而在提出的卷积自编码结构中的识别准确率达到了88.09%。实验结果表明,提出的算法在小样本图像识别中比卷积神经网络模型更加有效。     

遮挡表情变化下的联合辅助字典学习与低秩分解人脸识别

目的 从真实环境中采集到的人脸图片通常伴随遮挡、光照和表情变化等因素,对识别结果产生干扰。在许多特殊环境下,训练样本的采集数量也无法得到保证,容易产生训练样本远小于测试样本的不利条件。基于以上情况,如何排除复杂的环境变化和较少的训练样本等多重因素对识别效果的影响逐渐成为了人脸识别方向需要攻克的难题。方法 以低秩矩阵分解为基础,分别使用非凸秩近似范数和核范数进行两次低秩矩阵分解,以达到去除遮挡干扰的目的 。首先通过非凸稳健主成分分析分解得到去除了光照、遮挡等变化的低秩字典。为消除不同人脸类的五官等共通部分的影响,加快算法收敛效率,将得到的低秩字典用作初始化,进行基于核范数的第二次秩近似分解,以获得去除了类间不相关判别性的低秩字典用于分类。最后针对训练样本较少和遮挡样本占比过大等问题,选用同一数据库中不用做训练和测试的辅助数据作为辅助字典模拟可能出现的遮挡、光照等影响,通过最小化稀疏表示重构误差进行分类识别。结果 选用AR库和CK+库分别进行实验。在AR库的实验中,通过调整训练图片中遮挡、光照和表情变化的样本比例来检测算法性能。其中,在遮挡图片占比分别为1/7和3/7的训练集中,无遮挡图片由无干扰和光照表情干扰图片联合组成。在遮挡图片占比为2/7的训练集中,无遮挡图片全由光照表情变化图片组成。实验结果表明,在多种实验情况下均获得较高识别率。其中根据不同遮挡比例,分别获得97.75%、92%、95.25%和97.75%、90%、95.25%等识别率。与同类算法对比提高3%~5%。选用的外部数据从10类人脸至40类依次增加,获得的识别结果为96.75%~98%,与同类算法相比提高了2%~3%。在CK+表情库中,选用同伦算法配合分类求解,获得的识别结果为95.25%。结论 本文提出了一种在克服复杂环境变化和训练样本不足两个方面具有高效性和鲁棒性的人脸识别算法,实验结果表明,本文算法在不同数据库中都具有高效性,未来的研究方向包括将算法应用于联立人脸和表情识别,模拟更为复杂的噪声状况,以期达到更为优异的结果。     

Model-Driven Domain Adaptation on Product Manifolds for Unconstrained Face Recognition

Many classification algorithms see a reduction in performance when tested on data with properties different from that used for training. This problem arises very naturally in face recognition where images corresponding to the source domain (gallery, training data) and the target domain (probe, testing data) are acquired under varying degree of factors such as illumination, expression, blur and alignment. In this paper, we account for the domain shift by deriving a latent subspace or domain, which jointly characterizes the multifactor variations using appropriate image formation models for each factor. We formulate the latent domain as a product of Grassmann manifolds based on the underlying geometry of the tensor space, and perform recognition across domain shift using statistics consistent with the tensor geometry. More specifically, given a face image from the source or target domain, we first synthesize multiple images of that subject under different illuminations, blur conditions and 2D perturbations to form a tensor representation of the face. The orthogonal matrices obtained from the decomposition of this tensor, where each matrix corresponds to a factor variation, are used to characterize the subject as a point on a product of Grassmann manifolds. For cases with only one image per subject in the source domain, the identity of target domain faces is estimated using the geodesic distance on product manifolds. When multiple images per subject are available, an extension of kernel discriminant analysis is developed using a novel kernel based on the projection metric on product spaces. Furthermore, a probabilistic approach to the problem of classifying image sets on product manifolds is introduced. We demonstrate the effectiveness of our approach through comprehensive evaluations on constrained and unconstrained face datasets, including still images and videos.     

Pose-and-illumination-invariant face representation via a triplet-loss trained deep reconstruction model

Face recognition under variable pose and illumination is a challenging problem in computer vision tasks. In this paper, we solve this problem by proposing a new residual based deep face reconstruction neural network to extract discriminative pose-and-illumination-invariant (PII) features. Our deep model can change arbitrary pose and illumination face images to the frontal view with standard illumination. We propose a new triplet-loss training method instead of Euclidean loss to optimize our model, which has two advantages: a) The training triplets can be easily augmented by freely choosing combinations of labeled face images, in this way, overfitting can be avoided; b) The triplet-loss training makes the PII features more discriminative even when training samples have similar appearance. By using our PII features, we achieve 83.8% average recognition accuracy on MultiPIE face dataset which is competitive to the state-of-the-art face recognition methods.