基于视觉注意力机制的人脸属性迁移

人脸属性迁移作为计算机视觉领域的一个研究热点,对于数字娱乐制作、辅助人脸识别等领域有着重要的意义。现有的算法存在着生成图像模糊、转移属性无关区域变化等问题。针对这些不足,提出一种基于视觉注意力生成对抗网络的人脸属性迁移模型。生成器为减小属性无关区域的变化,引入视觉注意力分别输出RGB图像和注意力图像,并通过一定的融合方式得到属性迁移结果。采用多尺度判别器保持高维特征映射的细节。在约束中加入循环一致性损失和注意力图像损失,保持人脸身份信息,并专注属性相关区域的迁移。实验证明,该模型能够减少属性无关区域的变化,提高人脸属性转移的效果。     

2D representation of facial surfaces for multi-pose 3D face recognition

The increasing availability of 3D facial data offers the potential to overcome the intrinsic difficulties faced by conventional face recognition using 2D images. Instead of extending 2D recognition algorithms for 3D purpose, this letter proposes a novel strategy for 3D face recognition from the perspective of representing each 3D facial surface with a 2D attribute image and taking the advantage of the advances in 2D face recognition. In our approach, each 3D facial surface is mapped homeomorphically onto a 2D lattice, where the value at each site is an attribute that represents the local 3D geometrical or textural properties on the surface, therefore invariant to pose changes. This lattice is then interpolated to generate a 2D attribute image. 3D face recognition can be achieved by applying the traditional 2D face recognition techniques to obtained attribute images. In this study, we chose the pose invariant local mean curvature calculated at each vertex on the 3D facial surface to construct the 2D attribute image and adopted the eigenface algorithm for attribute image recognition. We compared our approach to state-of-the-art 3D face recognition algorithms in the FRGC (Version 2.0), GavabDB and NPU3D database. Our results show that the proposed approach has improved the robustness to head pose variation and can produce more accurate 3D multi-pose face recognition.     

MU-GAN: Facial Attribute Editing Based on Multi-Attention Mechanism

Facial attribute editing has mainly two objectives: 1) translating image from a source domain to a target one, and 2) only changing the facial regions related to a target attribute and preserving the attribute-excluding details. In this work, we propose a multi-attention U-Net-based generative adversarial network (MU-GAN). First, we replace a classic convolutional encoder-decoder with a symmetric U-Net-like structure in a generator, and then apply an additive attention mechanism to build attention-based U-Net connections for adaptively transferring encoder representations to complement a decoder with attribute-excluding detail and enhance attribute editing ability. Second, a self-attention (SA) mechanism is incorporated into convolutional layers for modeling long-range and multi-level dependencies across image regions. Experimental results indicate that our method is capable of balancing attribute editing ability and details preservation ability, and can decouple the correlation among attributes. It outperforms the state-of-the-art methods in terms of attribute manipulation accuracy and image quality. Our code is available at https://github.com/SuSir1996/MU-GAN.      

Mask guided diverse face image synthesis

Recent studies have shown remarkable success in face image generation task. However, existing approaches have limited diversity, quality and controllability in generating results. To address these issues, we propose a novel end-to-end learning framework to generate diverse, realistic and controllable face images guided by face masks. The face mask provides a good geometric constraint for a face by specifying the size and location of different components of the face, such as eyes, nose and mouse. The framework consists of four components: style encoder, style decoder, generator and discriminator. The style encoder generates a style code which represents the style of the result face; the generator translate the input face mask into a real face based on the style code; the style decoder learns to reconstruct the style code from the generated face image; and the discriminator classifies an input face image as real or fake. With the style code, the proposed model can generate different face images matching the input face mask, and by manipulating the face mask, we can finely control the generated face image. We empirically demonstrate the effectiveness of our approach on mask guided face image synthesis task.     

基于面部特征图对称的人脸正面化生成对抗网络算法

目前人脸正面化研究主要解决人脸偏转问题,而对监控视频等现实场景中同时受偏转和俯仰变化影响的侧脸的正面化生成关注较少,针对这个问题和多角度侧脸生成的正面人脸图存在身份信息保留不全的问题,提出了一种基于特征图对称模块和眼周特征保留损失的生成对抗网络（GAN）。首先,根据人脸对称性先验,提出特征图对称模块,先使用人脸关键点检测器检测出侧脸鼻尖点位置,再将编码器提取到的特征图依照鼻尖位置进行镜像对称,从而在特征层面上缓解面部信息缺失的问题。其次,借鉴眼周识别思想,在现有的生成图身份保留方法中加入了眼周特征保留损失以训练生成器生成逼真的且保留身份信息的人脸正面图像。实验结果表明,所提算法得到的生成图面部细节保留较好,且在CAS-PEAL-R1数据集的所有俯角下人脸的平均Rank-1识别率为99.03%,可见该算法能够有效解决多角度侧脸的正面化问题。     

A novel kernel-based framework for facial-image hallucination

In this paper, we present a kernel-based eigentransformation framework to hallucinate the high-resolution (HR) facial image of a low-resolution (LR) input. The eigentransformation method is a linear subspace approach, which represents an image as a linear combination of training samples. Consequently, those novel facial appearances not included in the training samples cannot be super-resolved properly. To solve this problem, we devise a kernel-based extension of the eigentransformation method, which takes higher-order statistics of the image data into account. To generate HR face images with higher fidelity, the HR face image reconstructed using this kernel-based eigentransformation method is treated as an initial estimation of the target HR face. The corresponding high-frequency components of this estimation are extracted to form a prior in the maximum a posteriori (MAP) formulation of the SR problem so as to derive the final reconstruction result. We have evaluated our proposed method using different kernels and configurations, and have compared these performances with some current SR algorithms. Experimental results show that our kernel-based framework, along with a proper kernel, can produce good HR facial images in terms of both visual quality and reconstruction errors.     

人脸老化/去龄化的高质量图像生成模型

目的 近年来关于人脸老化/去龄化的研究在深度学习的推动下取得了飞速发展,2017年提出的条件对抗自编码器（CAAE）人脸老化/去龄化模型生成的人脸不仅可信度高,而且更贴近目标年龄。然而在人脸老化/去龄化过程中仍存在生成图像分辨率低、人工鬼影噪声严重（生成人脸器官扭曲）等问题。为此,在CAAE的基础上,提出一个人脸老化/去龄化的高质量图像生成模型（HQGM）。方法 用边界平衡对抗生成网络（BEGAN）替换CAAE中的对抗生成网络（GAN）。BEGAN在人脸图像生成上不仅分辨率更高而且具有更好的视觉效果。在此基础上,添加两个提高生成图像质量的损失函数：图像梯度差损失函数和人脸特征损失函数。图像梯度差损失函数通过缩小生成图像和真实图像的图像梯度,使生成图像具有更多轮廓等高频信息;人脸特征损失函数将生成图像和真实图像分别输入到配置预训练参数的VGG-FACE网络模型中,输出各自的特征图。通过缩小两幅特征图的对应点差值,使生成图像具有更多真实图像的人脸特征信息。结果 实验使用UTKface、FGnet和Morph数据集,经过训练,每幅测试图像分别生成10幅不同年龄的图像。与CAAE相比,HQGM可以有效去除人工鬼影噪声,峰值信噪比高3.2 dB,结构相似性高0.06,提升显著。结论 HQGM可以生成具有丰富纹理信息和人脸特征信息的人脸老化/去龄化图像。     

基于条件生成式对抗网络的面部表情迁移模型

面部表情迁移是计算机视觉角色动画领域的关键技术,但现有面部表情迁移方法存在生成表情不自然、缺乏真实感、迁移模型复杂以及训练难度大等问题.为此,构建一种基于条件生成式对抗网络的面部表情迁移模型.通过设计域分类损失函数指定表情域条件,使单个生成器学习多个表情域之间的映射,同时利用模型生成器和判别器之间的条件约束与零和博弈,在仅训练一个生成器的情况下同时实现7种面部表情迁移.实验结果表明,该模型能够有效进行面部表情迁移并且鲁棒性较强,其生成的面部表情较StarGAN模型更自然、逼真.     

双重对偶生成对抗网络的跨年龄素描—照片转换

目的 跨年龄素描-照片转换旨在根据面部素描图像合成同一人物不同年龄阶段的面部照片图像。该任务在公共安全和数字娱乐等领域具有广泛的应用价值,然而由于配对样本难以收集和人脸老化机制复杂等原因,目前研究较少。针对此情况,提出一种基于双重对偶生成对抗网络（double dual generative adversarial networks,D-DualGANs）的跨年龄素描-照片转换方法。方法 该网络通过设置4个生成器和4个判别器,以对抗训练的方式,分别学习素描到照片、源年龄组到目标年龄组的正向及反向映射。使素描图像与照片图像的生成过程相结合,老化图像与退龄图像的生成过程相结合,分别实现图像风格属性和年龄属性上的对偶。并增加重构身份损失和完全重构损失以约束图像生成。最终使输入的来自不同年龄组的素描图像和照片图像,分别转换成对方年龄组下的照片和素描。结果 为香港中文大学面部素描数据集（Chinese University of Hong Kong（CUHK）face sketch database,CUFS）和香港中文大学面部素描人脸识别技术数据集（CUHK face sketch face recognition technology database,CUFSF）的图像制作对应的年龄标签,并依据标签将图像分成3个年龄组,共训练6个D-DualGANs模型以实现3个年龄组图像之间的两两转换。同非端到端的方法相比,本文方法生成图像的变形和噪声更小,且年龄平均绝对误差（mean absolute error,MAE）更低,与原图像相似度的投票对比表明1130素描与3150照片的转换效果最好。结论 双重对偶生成对抗网络可以同时转换输入图像的年龄和风格属性,且生成的图像有效保留了原图像的身份特征,有效解决了图像跨风格且跨年龄的转换问题。     

基于表情分解-扭曲变形的人工表情合成算法

为了能快速有效地生成任意强度的人脸表情图像，提出了一种鲁棒的可以生成带任意强度表情图像的人工表情合成算法，该算法首先通过施加高阶奇异值分解（HOSVD）来把训练集分解为个人、表情和特征3个子空间，并把它们映射到表情子空间中，用来合成任意人脸正面照片的任意强度、任意表情的图像；在生成图像时，不采用通常所使用的线性组合基图像生成法，而是对源图像进行扭曲变形，这不仅能使训练数据和计算量大为减少，还可以生成任意尺寸、任意背景、任意光照、任意色彩或任意姿势的表情图像，且通过二次插值，还可以得到任意强度的表情图像。实验证明，该算法效率较高，且生成的图像效果很好。     

人类面部重演方法综述

随着计算机视觉领域图像生成研究的发展,面部重演引起广泛关注,这项技术旨在根据源人脸图像的身份以及驱动信息提供的嘴型、表情和姿态等信息合成新的说话人图像或视频。面部重演具有十分广泛的应用,例如虚拟主播生成、线上授课、游戏形象定制、配音视频中的口型配准以及视频会议压缩等,该项技术发展时间较短,但是涌现了大量研究。然而目前国内外几乎没有重点关注面部重演的综述,面部重演的研究概述只是在深度伪造检测综述中以深度伪造的内容出现。鉴于此,本文对面部重演领域的发展进行梳理和总结。本文从面部重演模型入手,对面部重演存在的问题、模型的分类以及驱动人脸特征表达进行阐述,列举并介绍了训练面部重演模型常用的数据集及评估模型的评价指标,对面部重演近年研究工作进行归纳、分析与比较,最后对面部重演的演化趋势、当前挑战、未来发展方向、危害及应对策略进行了总结和展望。     

A video prediction approach for animating single face image

Generating dynamic 2D image-based facial expressions is a challenging task for facial animation. Much research work focused on performance-driven facial animation from given videos or images of a target face, while animating a single face image driven by emotion labels is a less explored problem. In this work, we treat the task of animating single face image from emotion labels as a conditional video prediction problem, and propose a novel framework by combining factored conditional restricted boltzmann machines (FCRBM) and reconstruction contractive auto-encoder (RCAE). A modified RCAE with an associated efficient training strategy is used to extract low dimensional features and reconstruct face images. FCRBM is used as animator to predict facial expression sequence in the feature space given discrete emotion labels and a frontal neutral face image as input. Both quantitative and qualitative evaluations on two facial expression databases, and comparison to state-of-the-art showed the effectiveness of our proposed framework for animating frontal neutral face image from given emotion labels.

     

基于图像的个性化卡通系统

介绍了一个卡通动画系统.该系统可使用户很方便地生成个性化的卡通头像,并通过简单的交互定制各种夸张而生动的表情,还可以实时生成语音驱动的卡通动画.系统主要包括3部分:基于图像的卡通自动生成模块、交互的卡通表情编辑模块和语音驱动的卡通动画模块.卡通自动生成模块采用基于样本学习的方法生成具有特定艺术风格的卡通头像.这里使用了一种非参数化采样算法来学习原始照片和画像间复杂的统计关系.在此基础上,卡通表情编辑器可以通过预先设计的模板生成各种具有生动表情的卡通.同时,用户也可以定制自己的表情模板.卡通动画模块使用了实时的唇形合成算法,可以生成语音驱动的卡通动画.该系统可以使普通用户很方便地生成个性化的卡通头像和动画,在这一点上优于其他动画制作系统.     

脸由音生：语音驱动的静动态人脸生成方法

语音驱动人脸生成旨在挖掘语音片段和人脸之间的静动态关联性,进而由给定的语音片段生成对应的人脸图像。然而已有的研究方法大多只考虑其中的一种关联性,且对静态人脸生成的研究严格依赖于时序对齐的音视频数据,在一定程度上限制了静态模型的使用范围。提出了一种基于条件生成对抗网络的语音驱动静动态人脸生成模型（SDVF-GAN）。该模型基于自注意力机制构建语音编码器网络以获得更为准确的听觉特征表达,并将其作为静态生成网络和动态生成网络的输入;静态生成网络利用基于投影层的图像判别器合成出属性一致（年龄、性别）且高质量的静态人脸图像,动态生成网络利用基于注意力思想的嘴唇判别器和图像判别器合成出嘴唇同步的动态人脸序列。实验利用所构建的属性对齐的Voice-Face数据集和公共的LRW数据集分别训练静态人脸生成网络和动态人脸生成网络。结果表明,该模型综合研究了语音和人脸之间的属性对应和嘴唇同步关系,实现了质量更高且关联性和同步性更强的人脸图像生成。     

选择性传输与铰链对抗的多图像域人脸属性迁移

在基于生成对抗网络的人脸属性迁移过程中,存在图像域表达形式单一、图像域迁移细节失真的问题。提出一种结合选择性传输单元与铰链对抗损失的多图像域人脸属性迁移方法。在生成器中,利用自适应实例归一化融合图像的内容信息与图像域控制器生成的样式信息,增加图像域表达方式的多样性,同时通过选择性传输单元将下采样提取的内容特征根据相对属性标签选择性地传输到上采样,形成融合特征以增强图像的细节信息。在判别器中,通过增加双尺度判别,协同鉴定人脸图像的真伪及类别,从而提高判定的准确度。在此基础上,设计融合相对鉴别和铰链损失的对抗损失函数,增强真伪图像域之间的联系。在CelebA数据集上的实验结果表明,与StarGAN、STGAN等主流的多图像域人脸属性迁移方法相比,该方法能够建立更准确的多图像域映射关系,提高迁移图像的质量同时增加迁移图像表达的多样性。     

基于性别和年龄因子分析的鲁棒性人脸表情识别

针对非可控环境下人脸表情识别面临的诸如种族、性别和年龄等因子变化问题,提出一种基于深度条件随机森林的鲁棒性人脸表情识别方法.与传统的单任务人脸表情识别方法不同,设计了一种以人脸表情识别为主,人脸性别和年龄属性识别为辅的多任务识别模型.在研究中发现,人脸性别和年龄等属性对人脸表情识别有一定的影响,为了捕获它们之间的关系,...     

Unified model in identity subspace for face recognition

Human faces have two important characteristics: (1) They are similar objects and the specific variations of each face are similar to each other; (2) They are nearly bilateral symmetric. Exploiting the two important properties, we build a unified model in identity subspace (UMIS) as a novel technique for face recognition from only one example image per person. An identity subspace spanned by bilateral symmetric bases, which compactly encodes identity information, is presented. The unified model, trained on an obtained training set with multiple samples per class from a known people group A, can be generalized well to facial images of unknown individuals, and can be used to recognize facial images from an unknown people group B with only one sample per subject. Extensive experimental results on two public databases (the Yale database and the Bern database) and our own database (the ICT-JDL database) demonstrate that the UMIS approach is significantly effective and robust for face recognition.     

基于Transformer实现文本导向的图像编辑

为方便非专业用户修图,提出一种基于Transformer的图像编辑模型TMGAN,使用户可通过自然语言描述自动修改图像属性。TMGAN整体框架采用生成对抗网络,生成器采用Transformer编码器结构提取全局上下文信息,解决生成图像不够真实的问题;判别器包含基于Transformer的多尺度判别器和词级判别器两部分,给生成器细粒度的反馈,生成符合文本描述的目标图像且保留原始图像中与文本描述无关的内容。实验表明,此模型在CUB Bird数据集上,IS（inception score）、FID（Fréchet inception distance）以及MP（manipulation precision）度量指标分别达到了9.07、8.64和0.081。提出的TMGAN模型对比现有模型效果更好,生成图像既满足了给定文本的属性要求又具有高语义性。