面部表情分析进展和挑战

面部表情分析是计算机通过分析人脸信息尝试理解人类情感的一种技术,目前已成为计算机视觉领域的热点话题。其挑战在于数据标注困难、多人标签一致性差、自然环境下人脸姿态大以及遮挡等。为了推动面部表情分析发展,本文概述了面部表情分析的相关任务、进展、挑战和未来趋势。首先,简述了面部表情分析的几个常见任务、基本算法框架和数据库;其次,对人脸表情识别方法进行了综述,包括传统的特征设计方法以及深度学习方法;接着,对人脸表情识别存在的问题与挑战进行总结思考;最后,讨论了未来发展趋势。通过全面综述和讨论,总结以下观点：1）针对可靠人脸表情数据库规模小的问题,从人脸识别模型进行迁移学习以及利用无标签数据进行半监督学习是两个重要策略;2）受模糊表情、低质量图像以及标注者的主观性影响,非受控自然场景的人脸表情数据的标签库存在一定的不确定性,抑制这些因素可以使得深度网络学习真正的表情特征;3）针对人脸遮挡和大姿态问题,利用局部块进行融合的策略是一个有效的策略,另一个值得考虑的策略是先在大规模人脸识别数据库中学习一个对遮挡和姿态鲁棒的模型,再进行人脸表情识别迁移学习;4）由于基于深度学习的表情识别方法受很多超参数影响,导致当前人脸表情识别方法的可比性不强,不同的表情识别方法有必要在不同的简单基线方法上进行评测。目前,虽然非受控自然环境下的表情分析得到较快发展,但是上述问题和挑战仍然有待解决。人脸表情分析是一个比较实用的任务,未来发展除了要讨论方法的精度也要关注方法的耗时以及存储消耗,也可以考虑用非受控环境下高精度的人脸运动单元检测结果进行表情类别推断。     

构建并行卷积神经网络的表情识别算法

目的 表情识别在商业、安全、医学等领域有着广泛的应用前景,能够快速准确地识别出面部表情对其研究与应用具有重要意义。传统的机器学习方法需要手工提取特征且准确率难以保证。近年来,卷积神经网络因其良好的自学习和泛化能力得到广泛应用,但还存在表情特征提取困难、网络训练时间过长等问题,针对以上问题,提出一种基于并行卷积神经网络的表情识别方法。方法 首先对面部表情图像进行人脸定位、灰度统一以及角度调整等预处理,去除了复杂的背景、光照、角度等影响,得到了精确的人脸部分。然后针对表情图像设计一个具有两个并行卷积池化单元的卷积神经网络,可以提取细微的表情部分。该并行结构具有3个不同的通道,分别提取不同的图像特征并进行融合,最后送入SoftMax层进行分类。结果 实验使用提出的并行卷积神经网络在CK+、FER2013两个表情数据集上进行了10倍交叉验证,最终的结果取10次验证的平均值,在CK+及FER2013上取得了94.03%与65.6%的准确率。迭代一次的时间分别为0.185 s和0.101 s。结论 为卷积神经网络的设计提供了一种新思路,可以在控制深度的同时扩展广度,提取更多的表情特征。实验结果表明,针对数量、分辨率、大小等差异较大的表情数据集,该网络模型均能够获得较高的识别率并缩短训练时间。     

结合形变模型与图像修复的人脸姿态矫正

目的 人脸姿态偏转是影响人脸识别准确率的一个重要因素,本文利用3维人脸重建中常用的3维形变模型以及深度卷积神经网络,提出一种用于多姿态人脸识别的人脸姿态矫正算法,在一定程度上提高了大姿态下人脸识别的准确率。方法 对传统的3维形变模型拟合方法进行改进,利用人脸形状参数和表情参数对3维形变模型进行建模,针对面部不同区域的关键点赋予不同的权值,加权拟合3维形变模型,使得具有不同姿态和面部表情的人脸图像拟合效果更好。然后,对3维人脸模型进行姿态矫正并利用深度学习对人脸图像进行修复,修复不规则的人脸空洞区域,并使用最新的局部卷积技术同时在新的数据集上重新训练卷积神经网络,使得网络参数达到最优。结果 在LFW（labeled faces in the wild）人脸数据库和StirlingESRC（Economic Social Research Council）3维人脸数据库上,将本文算法与其他方法进行比较,实验结果表明,本文算法的人脸识别精度有一定程度的提高。在LFW数据库上,通过对具有任意姿态的人脸图像进行姿态矫正和修复后,本文方法达到了96.57%的人脸识别精确度。在StirlingESRC数据库上,本文方法在人脸姿态为±22°的情况下,人脸识别准确率分别提高5.195%和2.265%;在人脸姿态为±45°情况下,人脸识别准确率分别提高5.875%和11.095%;平均人脸识别率分别提高5.53%和7.13%。对比实验结果表明,本文提出的人脸姿态矫正算法有效提高了人脸识别的准确率。结论 本文提出的人脸姿态矫正算法,综合了3维形变模型和深度学习模型的优点,在各个人脸姿态角度下,均能使人脸识别准确率在一定程度上有所提高。     

深度人脸表情识别研究进展

随着人脸表情识别任务逐渐从实验室受控环境转移至具有挑战性的真实世界环境,在深度学习技术的迅猛发展下,深度神经网络能够学习出具有判别能力的特征,逐渐应用于自动人脸表情识别任务。目前的深度人脸表情识别系统致力于解决以下两个问题：1）由于缺乏足量训练数据导致的过拟合问题;2）真实世界环境下其他与表情无关因素变量（例如光照、头部姿态和身份特征）带来的干扰问题。本文首先对近十年深度人脸表情识别方法的研究现状以及相关人脸表情数据库的发展进行概括。然后,将目前基于深度学习的人脸表情识别方法分为两类：静态人脸表情识别和动态人脸表情识别,并对这两类方法分别进行介绍和综述。针对目前领域内先进的深度表情识别算法,对其在常见表情数据库上的性能进行了对比并详细分析了各类算法的优缺点。最后本文对该领域的未来研究方向和机遇挑战进行了总结和展望：考虑到表情本质上是面部肌肉运动的动态活动,基于动态序列的深度表情识别网络往往能够取得比静态表情识别网络更好的识别效果。此外,结合其他表情模型如面部动作单元模型以及其他多媒体模态,如音频模态和人体生理信息能够将表情识别拓展到更具有实际应用价值的场景。     

采用深度学习方法的非正面表情识别综述

2017年人工智能正式升级为中国国家战略,作为人工智能领域中重要的研究方向,人脸表情识别受到了国内外研究者们的广泛关注。然而传统的人脸表情识别技术无法适应自然环境下的表情识别需求。因此非正面人脸表情识别方法成为实现表情识别技术实用化突破的重点。但是现有的非正面表情识别研究面临很多困难：头部偏转不仅造成了识别图像的扭曲,而且还遮挡了部分人脸区域,严重干扰了表情特征的提取与识别。有鉴于此,研究者们将深度学习技术与非正面表情识别相结合,依靠非正面表情图像的深度信息,实现算法识别能力的提升。综述详细介绍了深度神经网络的结构,对最新的深度学习神经网络研究方法进行分类对比,同时对未来的研究和挑战做了展望。     

面向大姿态人脸识别的正面化形变场学习

目的 人脸识别已经得到了广泛应用,但大姿态人脸识别问题仍未完美解决。已有方法或提取姿态鲁棒特征,或进行人脸姿态的正面化。其中主流的人脸正面化方法包括2D回归生成和3D模型形变建模,前者能够生成相对自然真实的人脸,但会引入额外的噪声导致图像信息的扭曲;后者能够保持原始的人脸结构信息,但生成过程是基于物理模型的,不够自然灵活。为此,结合2D和3D方法的优势,本文提出了基于由粗到细形变场的人脸正面化方法。方法 该形变场由深度网络以2D回归方式学得,反映的是不同视角人脸图像像素之间的语义级对应关系,可以类3D的方式实现非正面人脸图像的正面化,因此该方法兼具了2D正面化方法的灵活性与3D正面化方法的保真性,且借鉴分步渐进的思路,本文提出了由粗到细的形变场学习框架,以获得更加准确鲁棒的形变场。结果 本文采用大姿态人脸识别实验来验证本文方法的有效性,在MultiPIE(multi pose, illumination, expressions)、LFW(labeled faces in the wild)、CFP(celebrities in frontal-profile in the wild)...     

Automatic extraction of eye and mouth fields from a face image using eigenfeatures and multilayer perceptrons

This paper presents a novel algorithm for the extraction of the eye and mouth (facial features) fields from 2-D gray-level face images. The fundamental philosophy is that eigenfeatures, derived from the eigenvalues and eigenvectors of the binary edge data set constructed from the eye and mouth fields, are very good features to locate these fields efficiently. The eigenfeatures extracted from the positive and negative training samples of the facial features are used to train a multilayer perceptron whose output indicates the degree to which a particular image window contains an eye or a mouth. It turns out that only a small number of frontal faces are sufficient to train the networks. Furthermore, they lend themselves to good generalization to non-frontal pose and even other people's faces. It has been experimentally verified that the proposed algorithm is robust against facial size and slight variations of pose.     

Pose Robust Low-resolution Face Recognition via Coupled Kernel-based Enhanced Discriminant Analysis

Most face recognition techniques have been successful in dealing with high-resolution (HR) frontal face images. However, real-world face recognition systems are often confronted with the low-resolution (LR) face images with pose and illumination variations. This is a very challenging issue, especially under the constraint of using only a single gallery image per person. To address the problem, we propose a novel approach called coupled kernel-based enhanced discriminant analysis (CKEDA). CKEDA aims to simultaneously project the features from LR non-frontal probe images and HR frontal gallery ones into a common space where discrimination property is maximized. There are four advantages of the proposed approach: 1) by using the appropriate kernel function, the data becomes linearly separable, which is beneficial for recognition; 2) inspired by linear discriminant analysis (LDA), we integrate multiple discriminant factors into our objective function to enhance the discrimination property; 3) we use the gallery extended trick to improve the recognition performance for a single gallery image per person problem; 4) our approach can address the problem of matching LR non-frontal probe images with HR frontal gallery images, which is difficult for most existing face recognition techniques. Experimental evaluation on the multi-PIE dataset signifies highly competitive performance of our algorithm.      

融合时空域特征的人脸表情识别

目的 人脸表情识别是计算机视觉的核心问题之一。一方面,表情的产生对应着面部肌肉的一个连续动态变化过程,另一方面,该运动过程中的表情峰值帧通常包含了能够识别该表情的完整信息。大部分已有的人脸表情识别算法要么基于表情视频序列,要么基于单幅表情峰值图像。为此,提出了一种融合时域和空域特征的深度神经网络来分析和理解视频序列中的表情信息,以提升表情识别的性能。方法 该网络包含两个特征提取模块,分别用于学习单幅表情峰值图像中的表情静态“空域特征”和视频序列中的表情动态“时域特征”。首先,提出了一种基于三元组的深度度量融合技术,通过在三元组损失函数中采用不同的阈值,从单幅表情峰值图像中学习得到多个不同的表情特征表示,并将它们组合在一起形成一个鲁棒的且更具辩识能力的表情“空域特征”;其次,为了有效利用人脸关键组件的先验知识,准确提取人脸表情在时域上的运动特征,提出了基于人脸关键点轨迹的卷积神经网络,通过分析视频序列中的面部关键点轨迹,学习得到表情的动态“时域特征”;最后,提出了一种微调融合策略,取得了最优的时域特征和空域特征融合效果。结果 该方法在3个基于视频序列的常用人脸表情数据集CK+(the e...     

持续学习改进的人脸表情识别

目的 大量标注数据和深度学习方法极大地提升了图像识别性能。然而,表情识别的标注数据缺乏,训练出的深度模型极易过拟合,研究表明使用人脸识别的预训练网络可以缓解这一问题。但是预训练的人脸网络可能会保留大量身份信息,不利于表情识别。本文探究如何有效利用人脸识别的预训练网络来提升表情识别的性能。方法 本文引入持续学习的思想,利用人脸识别和表情识别之间的联系来指导表情识别。方法指出网络中对人脸识别整体损失函数的下降贡献最大的参数与捕获人脸公共特征相关,对表情识别来说为重要参数,能够帮助感知面部特征。该方法由两个阶段组成：首先训练一个人脸识别网络,同时计算并记录网络中每个参数的重要性;然后利用预训练的模型进行表情识别的训练,同时通过限制重要参数的变化来保留模型对于面部特征的强大感知能力,另外非重要参数能够以较大的幅度变化,从而学习更多表情特有的信息。这种方法称之为参数重要性正则。结果 该方法在RAF-DB（real-world affective faces database）,CK+（the extended Cohn-Kanade database）和Oulu-CASIA这3个数据集上进行了实验评估。在主流数据集RAF-DB上,该方法达到了88.04%的精度,相比于直接用预训练网络微调的方法提升了1.83%。其他数据集的实验结果也表明了该方法的有效性。结论 提出的参数重要性正则,通过利用人脸识别和表情识别之间的联系,充分发挥人脸识别预训练模型的作用,使得表情识别模型更加鲁棒。     

基于不同视角的人脸模型识别方法

对在不同视角下,得到的人脸模型,文中提出一种基于人脸表面的识别方法,采用平面射影变换,将人脸的图像变换到一个相同的图像,使图像对齐;而后运用主成分分析法(PCA)进行分类。基于这种方法,由光线、面部表情、姿态的改变引起的不必要变化就可能被消除或可以忽略。这种方法可以达到比较准确的识别人脸的目的。实验结果显示,文中的方法对人脸模型提供了更好的表达,并且人脸识别的错误率更低。     

Learning warps based similarity for pose-unconstrained face recognition

Face recognition techniques are widely used in many applications, such as automatic detection of crime scenes from surveillance cameras for public safety. In these real cases, the pose and illumination variances between two matching faces have a big influence on the identification performance. Handling pose changes is an especially challenging task. In this paper, we propose the learning warps based similarity method to deal with face recognition across the pose problem. Warps are learned between two patches from probe faces and gallery faces using the Lucas-Kanade algorithm. Based on these warps, a frontal face registered in the gallery is transformed into a series of non-frontal viewpoints, which enables non-frontal probe face matching with the frontal gallery face. Scale-invariant feature transform (SIFT) keypoints (interest points) are detected from the generated viewpoints and matched with the probe faces. Moreover, based on the learned warps, the probability likelihood is used to calculate the probability of two faces being the same subject. Finally, a hybrid similarity combining the number of matching keypoints and the probability likelihood is proposed to describe the similarity between a gallery face and a probe face. Experimental results show that our proposed method achieves better recognition accuracy than other algorithms it was compared to, especially when the pose difference is within 40 degrees.

     

A novel 2D and 3D multimodal approach for in-the-wild facial expression recognition

This study proposes a novel deep learning approach for the fusion of 2D and 3D modalities in in-the-wild facial expression recognition (FER). Different from other studies, we exploit the 3D facial information in in-the-wild FER. In particular, in-the-wild 3D FER dataset is not widely available; therefore, 3D facial data are constructed from available 2D datasets thanks to recent advances in 3D face reconstruction. The 3D facial geometry features are then extracted by deep learning technique to exploit the mid-level details, which provides meaningful expression for the recognition. In addition, to demonstrate the potential of 3D data on FER, the 2D projected images of 3D faces are taken as additional input to FER. These features are then jointly fused with 2D features obtained from the original input. The fused features are then classified by support vector machines (SVMs). The results show that the proposed approach achieves state-of-the-art recognition performances on Real-World Affective Faces (RAF) and Static Facial Expressions in the Wild (SFEW 2.0), and AffectNet dataset. This approach is also applied to a 3D FER dataset, i.e. BU-3DFE, to compare the effectiveness of reconstructed and available 3D face data for FER. This is the first time such a deep learning combination of 3D and 2D facial modalities is presented in the context of in-the-wild FER.     

Higher order orthogonal moments for invariant facial expression recognition

Automatic facial expression recognition (FER) is a sub-area of face analysis research that is based heavily on methods of computer vision, machine learning, and image processing. This study proposes a rotation and noise invariant FER system using an orthogonal invariant moment, namely, Zernike moments (ZM) as a feature extractor and Naive Bayesian (NB) classifier. The system is fully automatic and can recognize seven different expressions. Illumination condition, pose, rotation, noise and others changing in the image are challenging task in pattern recognition system. Simulation results on different databases indicated that higher order ZM features are robust in images that are affected by noise and rotation, whereas the computational rate for feature extraction is lower than other methods.     

深度学习下融合不同模型的小样本表情识别

为了进一步提高人脸表情识别在小样本中的准确率,提出了一种深度学习下融合不同模型的小样本表情识别方法。该方法首先对单个卷积神经网络(CNN)模型进行比较,通过dropout层不同的节点保留概率p,筛选相对合适的CNN。之后采用尺度不变特征变换(SIFT)算法提取出特征,使用SIFT提取特征的目的是提高小数据的性能。为了减少误差,避免过拟合,将所有模型进行汇总,采用简单平均的模型融合方法得到CNN-SIFT-AVG模型。最后,只采用少量样本数据来训练模型即可。该模型已在FER2013、CK+和JAFFE数据集上进行了验证实验。实验结果表明,该模型可以很大程度上提高小样本表情识别的准确率,并在FER2013、CK+和JAFFE数据集上产生了较优异的结果,与其他表情识别方法相比,准确率最大提升约6%。     

自适应监督下降方法的姿态鲁棒人脸对齐算法

人脸对齐是人脸分析处理中的重要一步。由于现实中的人脸照片通常在姿态、光线等方面存在较大的差异,人脸对齐是一项艰巨的任务。初始关键点的位置以及特征提取对人脸对齐很重要。提出一种自适应监督下降方法(SDM)的姿态鲁棒人脸对齐算法。首先,为了减小姿态差异对人脸对齐的影响,使用聚类算法将图片按照姿态分成三类(正脸,左侧脸,右侧脸),这样每个类别下的姿态更加紧致。其次,考虑到人脸对齐是由粗到细的多阶段监督学习过程,采用自适应特征提取框(由大到小)来提取判别性特征。基于上述两种策略,在每个类别下,提供一个更好的初始关键点位置,通过自适应特征提取的SDM模型来进行回归模型的训练。选用LFPW、HELEN和300W数据集进行评估,实验结果表明,该模型在复杂姿态下能准确定位关键点,并且好于现有的人脸对齐算法。     

Pose invariant face recognition based on hybrid-global linear regression

The paper presents a simple but efficient novel H-eigenface (Hybrid-eigenface) method for pose invariant face recognition ranging from frontal to profile view. H-eigenfaces are entirely new basis for face image representation under different poses and are used for virtual frontal view synthesis. The proposed method is based on the fact that face samples of same person under different poses are similar in terms of the combination pattern of facial features. H-eigenfaces exploit this fact and thus two H-eigenfaces under different poses capture same features of the face. Thereby providing a compact view-based subspace, which can be further used to generate virtual frontal view from inputted non-frontal face image using least square projection technique. The use of proposed methodology on FERET and ORL face database shows an impressive improvement in recognition accuracy and a distinct reduction in online computation when compared to global linear regression method.     

基于多任务学习的多姿态人脸重建与识别

针对当前人脸识别中姿态变化会影响识别性能,以及姿态恢复过程中脸部局部细节信息容易丢失的问题,提出一种基于多任务学习的多姿态人脸重建与识别方法——多任务学习堆叠自编码器（MtLSAE）。该方法通过运用多任务学习机制,联合考虑人脸姿态恢复和脸部局部细节信息保留这两个相关的任务,在步进逐层恢复正面人脸姿态的同时,引入非负约束稀疏自编码器,使得非负约束稀疏自编码器能够学习到人脸部的部分特征;其次在姿态恢复和局部信息保留两个任务之间通过共享参数的方式来学习整个网络框架;最后将重建出来的正脸图像通过Fisherface进行降维并提取具有判别信息的特征,并用最近邻分类器来识别。实验结果表明,MtLSAE方法获得了较好的姿态重建质量,保留的局部纹理信息清晰,而且与局部Gabor二值模式（LGBP）、基于视角的主动外观模型（VAAM）以及堆叠步进自编码器（SPAE）等经典方法相比,识别率性能得以提升。     

Automatic interpretation and coding of face images using flexiblemodels

Face images are difficult to interpret because they are highly variable. Sources of variability include individual appearance, 3D pose, facial expression, and lighting. We describe a compact parametrized model of facial appearance which takes into account all these sources of variability. The model represents both shape and gray-level appearance, and is created by performing a statistical analysis over a training set of face images. A robust multiresolution search algorithm is used to fit the model to faces in new images. This allows the main facial features to be located, and a set of shape, and gray-level appearance parameters to be recovered. A good approximation to a given face can be reconstructed using less than 100 of these parameters. This representation can be used for tasks such as image coding, person identification, 3D pose recovery, gender recognition, and expression recognition. Experimental results are presented for a database of 690 face images obtained under widely varying conditions of 3D pose, lighting, and facial expression. The system performs well on all the tasks listed above