认知与发展心理学研究室知识库

欺骗行为普遍存在于日常生活中。人们通过隐藏真实信息或故意提供虚假信息来获取更多生存资源，但受骗方往往因此而利益受损。积极探索提高欺骗检测准确率的方法具有现实意义。人的面孔能够传递诸多信息，通过面孔读取他人心理活动的行为由来已久。人们习惯通过快速“解读”面部信息来评价一个陌生人，并通过这些面部信息评估他们所处的心理状态，解读面部信息可以为了解他人意图、动机和交流，甚至辨别欺骗提供重要线索。本研究主要探讨了欺骗检测中有效的面部行为线索，通过研究人们进行面孔特征解读时的信任判断偏好发现潜在的欺骗检测面部线索并在模拟犯罪实验中进行验证，同时为了更好地应用于实际，本研究还对如何提高面部线索的编码效率进行探索，具体通过下面三个研究进行：

研究一（实验1和实验2）考察面孔特征的信任判断偏好。实验1中采用CASME Ⅱ微表情数据库中的图片，要求被试观看不同微表情类型的面孔图片序列，对信任程度做出评分。结果表明，人们在解读他人面孔特征时表现出特定的信任判断偏好：具有恐惧和紧张微表情的面孔容易让人产生怀疑和不信任，而具有高兴、悲伤和惊讶微表情的面孔容易取得人的信任，厌恶微表情的面孔没有特定信任偏好。针对实验1中负面信任判断偏好的微表情面孔均具有巩膜暴露面积变小的特征，实验2从眼睛的瞳孔和巩膜不同暴露比例进一步探索，通过改变眼睛的黑白比例（是否佩戴美瞳）来探索眼睛的白色巩膜暴露程度是否会影响解读面孔特征时的信任判断偏好。结果表明，人们对于有更多白色巩膜暴露的人表现出更多的不信任。

研究二（实验3和实验4）探索模拟犯罪情境下欺骗检测的面部线索。实验3通过模拟犯罪室的偷窃任务和模拟审讯室的欺骗检测任务对研究一人们解读面孔特征的信任判断偏好是否能作为欺骗检测的面部线索进行验证。结果表明，人们在进行欺骗行为时会产生一定的面部线索，但是这些线索具有个体差异，并且有些面部线索的实际表达和人们的信任判断偏好不一致。高兴作为人们普遍信任的面孔特征，却是很多欺骗者喜欢采用的。实验4在实验3的基础上增加了欺骗风险程度的分级，通过偷窃假装没偷和没偷假装偷窃来模拟两种欺骗风险水平下的面部线索表达。结果表明，高低不同风险环境下，欺骗检测面部线索中宏表情的表达没有显著差异，但微表情的表达数量在高风险环境时要多于低风险环境，在表现类型上，低风险水平中宏表情和微表情出现的类型一致，但高风险水平下两者出现的类型不一样，宏表情中出现最多的类型是恐惧和惊讶，微表情中出现最多的类型是高兴。

研究三探索面部线索的高效编码。现阶段面部运动信息编码效率的低下限制了其在生活中的推广，本研究为了推动面部线索在欺骗检测中的实际应用，对提高面部运动信息编码效率做了积极探索：首先，从自动编码角度，为训练微表情高效编码的算法，构建了一个带有深度信息的微表情数据库以提供可靠样本。然后，从人工编码角度，开发两款微表情标注软件并搭建一个多人在线编码平台——微表情编码与分享系统，改善了人工编码环境。这些研究成果对提高面部运动的编码效率起到了积极作用。

本研究发现了人们在试图解读他人面部信息时的微表情特征判断偏好，通过模拟犯罪实验对欺骗检测的面部线索进行验证，开发面部信息编码的辅助工具和网站，有助于推动面部线索在欺骗检测中的实际应用。

Deceptive behavior is common in daily life. People get more living resources by hiding the real information or deliberately providing false information, but the cheated party often suffers from the loss of interests. It is of practical significance to actively explore ways to improve the accuracy of deception detection. There are many colors of facial makeup in Peking Opera, and the different colors usually have specific meanings: red represents loyalty and chivalry, black represents straightforwardness and fortitude, and white represents treachery. Reading information from other people's faces has a long history. People are used to quickly "reading" facial information to judge a stranger and assess their state of mind. Reading facial information can provide important clues to other people's intentions, motivations, and communication, and even detect deception. For nearly half a century, people have believed that psychological states and the nature of personality can be inferred from facial features. This research mainly probes into the effective behavior of facial clues in deception detection, by the people for the interpretation of the information of the facial preferences found the potential for fraud detection facial cues and in simulated experiments validated the crime, to better applied to the actual at the same time, this study is to improve the coding efficiency of facial clues to explore, research on concrete from the following three:

Study 1 (experiment 1, experiment 2) examined the reading preference of facial cues. In experiments 1, stimulus were taken from database CASME Ⅱ, the participants evaluated face image sequence with different types of micro-expression. We mainly selected six types of prime-peak-end frames of happiness, disgust, sadness, fear, surprise, and tension as a group of images. The results of experiment 1 showed that people had certain preferences when reading other people's faces with different micro-expression: fear and nervous faces were more likely to raise doubts, while happy, sad, and surprised faces were more likely to be trusted, and disgust face showed no significant preference. Experiment 2 studied the reading preference in different proportion of pupil and sclera, because of the fear and nervous face contained frown movement, this facial movement led to the eyelids cover the black pupil of the eye, in addition, to change shape, “black eyes” and “white eyes” exposed also decreases, so we explored the white sclera in facial reading preferences by changing the proportion of black and white of the eye (wore lenses or not). The results of experiment 2 showed that people had more distrust of people with more white sclera exposure.

Study 2 (experiment 3, experiment 4) explored the facial clues of deception detection in mock crime. In experiment 3, there were two parts, mock thefts, and mock trials. During the experiment, all the behaviors were determined by the participants, and the high bonus was regarded as the high motivation to deception. The results of experiment 3 showed that when people chose to cheat, they would generate certain facial cues, but these cues have individual differences, and some facial cues were inconsistent with people's reading preference. Happy was generally trusted face by most people, but it was usually produced after deception. Based on experiment 3, experiment 4 increased the low level of deception risk. In the two levels of deception risk, facial cue expression was simulated by pretending to steal and pretending not to steal. Other experimental processes were the same as experiment 3. The results of experiment 4 showed that there was no significant difference in macro expressions as facial cues of deception detection in high and low-risk environments, but the micro-expression at high-risk levels was much than that at a low-risk level. In high level, macro-expressions showed the most fear and surprise, while micro-expression showed the most happiness.

Study 3 efficient coding of facial cues. To promote the practical application of facial cues in deception detection, this study made three active explorations in improving the coding of facial movement: constructing a micro-expression database with deep information, developing two micro-expression coding software and a multi-person online coding platform, micro-expression coding and sharing system. These performances have greatly improved the coding efficiency of facial cues.

This study found people's trust preference when trying to read other people's facial information, verified the facial clues of deception detection through mock crime experiments. We have also developed auxiliary tools and websites for coding facial action information, which can help promote the practical application of deception detection.