面部运动单元检测研究综述

面部运动单元检测旨在让计算机从给定的人脸图像或视频中自动检测需要关注的运动单元目标。经过二十多年的研究，尤其是近年来越来越多的面部运动单元数据库的建立和深度学习的兴起，面部运动单元检测技术发展迅速。首先，阐述了面部运动单元的基本概念，介绍了已有的常用面部运动单元检测数据库，概括了包括预处理、特征提取、分类器学习等步骤在内的传统检测方法；然后针对区域学习、面部运动单元关联学习、弱监督学习等几个关键研究方向进行了系统性的回顾梳理与分析；最后讨论了目前面部运动单元检测研究存在的不足以及未来潜在的发展方向。     

Attentional Biases for Negative Interpersonal Stimuli in Clinical Depression.

An information-processing paradigm was used to examine attentional biases in clinically depressed participants, participants with generalized anxiety disorder (GAD), and nonpsychiatric control participants for faces expressing sadness, anger, and happiness. Faces were presented for 1,000 ms, at which point depressed participants had directed their attention selectively to depression-relevant (i.e., sad) faces. This attentional bias was specific to the emotion of sadness; the depressed participants did not exhibit attentional biases to the angry or happy faces. This bias was also specific to depression; at 1,000 ms, participants with GAD were not attending selectively to sad, happy, or anxiety-relevant (i.e., angry) faces. Implications of these findings for both the cognitive and the interpersonal functioning of depressed individuals are discussed and directions for future research are advanced. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

PS,LS和煤油体系的界面张力

本文以石油磺酸盐(PS)、木质素磺酸盐(LS)和煤油体系为对象,研究了PS/LS比、盐度、醇、Na_2SiO_3及聚合物对体系界面张力(IFT)的影响,探讨了PS与各组分间的相互作用,结果发现,在一定盐度和混合醇(正丁醇+异丙醇)浓度下,PS/LS>0.3/0.7时,体系可形成中相微乳液,IFT值达1×10~(-3)mN/m左右;LS和聚合物的存在使IFT升高;在表面活性剂驱油体系中,可用廉价的LS部分代替PS。     

Contributions of Oral and Extraoral Facial Movement to Visual and Audiovisual Speech Perception.

Seeing a talker's face influences auditory speech recognition, but the visible input essential for this influence has yet to be established. Using a new seamless editing technique, the authors examined effects of restricting visible movement to oral or extraoral areas of a talking face. In Experiment 1, visual speech identification and visual influences on identifying auditory speech were compared across displays in which the whole face moved, the oral area moved, or the extraoral area moved. Visual speech influences on auditory speech recognition were substantial and unchanging across whole-face and oral-movement displays. However, extraoral movement also influenced identification of visual and audiovisual speech. Experiments 2 and 3 demonstrated that these results are dependent on intact and upright facial contexts, but only with extraoral movement displays. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

基于Eigenspace的面瘫判定系统

本文将特征空间分类方法应用于面瘫图像的判定。结合应用的特点，引入多特征空间分类，比较图像与它们到两个特征空间的投影之间的距离作出判定，并在系统中加入灰度图变换，脸部子区域分割，确定对称轴和输入向量计算等前处理步骤。实验证明，本文的改进提高了系统判定的正确性。     

Suboptimal Exposure to Facial Expressions When Viewing Video Messages From a Small Screen: Effects on Emotion, Attention, and Memory.

The authors examined the effects of suboptimally presented facial expressions on emotional and attentional responses and memory among 39 young adults viewing video (business news) messages from a small screen. Facial electromyography (EMG) and respiratory sinus arrhythmia were used as physiological measures of emotion and attention, respectively. Several congruency priming effects were found. In particular, happy facial primes prompted increased (a) pleasure ratings, (b) orbicularis oculi EMG activity, (c) perceived trustworthiness, and (d) recognition memory for video messages with a positive emotional tone. Emotional and other responses to video messages presented on a small screen can be modified with suboptimal affective primes, but even small differences in the emotional tone of the messages should be allowed for. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Garner interference reveals dependencies between emotional expression and gaze in face perception.

The relationship between facial expression and gaze processing was investigated with the Garner selective attention paradigm. In Experiment 1, participants performed expression judgments without interference from gaze, but expression interfered with gaze judgments. Experiment 2 replicated these results across different emotions. In both experiments, expression judgments occurred faster than gaze judgments, suggesting that expression was processed before gaze could interfere. In Experiments 3 and 4, the difficulty of the emotion discrimination was increased in two different ways. In both cases, gaze interfered with emotion judgments and vice versa. Furthermore, increasing the difficulty of the emotion discrimination resulted in gaze and expression interactions. Results indicate that expression and gaze interactions are modulated by discriminability. Whereas expression generally interferes with gaze judgments, gaze direction modulates expression processing only when facial emotion is difficult to discriminate. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The timing of appraisals.

The appraisal process consists of the subjective evaluation that occurs during an individual's encounter with significant events in the environment, determining the nature of the emotional reaction and experience. Placed in the context of appraisal theories of emotion-elicitation and differentiation, the aim of the present research was to test empirically the hypothesis that the intrinsic pleasantness evaluation occurs before the goal conduciveness evaluation. In two studies, intrinsically pleasant and unpleasant images were used to manipulate pleasantness, and a specific event in a Pacman-type videogame was used to manipulate goal conduciveness. Facial EMG was used to measure facial reactions to each evaluation. As predicted, facial reactions to the intrinsic pleasantness manipulation were faster than facial reactions to the goal conduciveness manipulation. These results provide good empirical support for the sequential nature of the appraisal process. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

基于多任务学习和知识图谱的面部表情识别

针对面部表情分类的模型中参数较复杂、识别准确率较低的问题,提出了一种基于知识图谱辅助识别的多任务学习算法模型（MLAM）,该模型由基于深度学习的识别模块与知识图谱嵌入模块两部分构成。首先从输入的数据中提取潜在的人脸局部表情特征,通过知识图谱实现局部表情和个体的复杂交互;然后在MLAM 模型中设计一个交叉压缩单元,关联这两个独立模块,自动学习局部表情和实体特征的高级交互,并在这两个任务之间传递交叉知识转移;最后,在FER2013和CK+的数据集上对比了同类算法,实验结果表明,该模型在上述数据集上分别得到了0.69和0.99的识别率,提高了面部表情识别准确率。     

基于人脸特征和线积分卷积的肖像素描生成

为了更好地模拟素描的笔画和突显肖像素描中五官的部分,提出基于人脸特征和线积分卷积的肖像素描生成方法.首先进行人脸分割,由人脸特征点确定GrabCut算法中每个特征的初始trimap,然后根据分割结果使用线积分卷积算法绘制肖像素描.文中使用了基于分割结果的多分辨率白噪声和方向场,并提出了基于分割结果的自适应性边界提取算法.实验结果表明,该算法改善了肖像素描的生成效果.