基于稀疏张量的彩色微表情识别
详细信息 查看官网全文
- 英文论文题名：Color Micro-expression Recognition based on Sparse Tensor
- 论文作者：王甦菁
- 英文论文作者：Su-Jing Wang ; Key Laboratory of Behavioral Sciences ; Chinese Academy of Sciences
- 年：2015
- 作者机构：中国科学院行为科学重点实验室;
- 论文关键词：微表情识别 ; 颜色空间 ; 张量分析 ; 稀疏表示
- 英文论文关键词：Micro-expression recognition ; Color space ; Tensor analysis ; Sparse representation
- 会议召开时间：2015-10-16
- 会议录名称：第十八届全国心理学学术会议摘要集——心理学与社会发展
- 英文会议录名称：Proceedings of the 18th National Academic Congress of Psychology
- 语种：中文
- 分类号：B842
- 学会代码：ZGXG
- 会议名称：第十八届全国心理学学术会议
- 会议地点：中国天津
- 主办单位：中国心理学会
- 学会名称：中国心理学会
- 页数：2
- 文件大小：740k
- 原文格式：D
- 会议级别：全国
摘要
微表情是一种快速泄露的,其特点:持续时间短变化幅度小。它在自动谎言识别等众多领域都有谎言识别等众多领域都有巨大的应用价值。本研究综合使计算机视觉技术与认知心理学实验方法,研究微表情自动识别算及模型。构建微表情数据库。研究彩色空间,用彩色线索进一步提高微表情的识别率,针对微表情的特点,研究微表情的稀疏表示,并结合张量的分析来保持其空间结构信息。具体来说:1.我们通过心理学的手段诱发出微表情,并用高速摄像机进行采集。构建并发布的两个表情的数据库。2.我们扩展判别张量子空间分析到高阶张量上并使用极限学习机做分类器,对灰度微表情视频进行识别。3.研究彩色空间,提出一个新的颜色空间模型,张量独立彩色空间(TICS)。在TICS中,微表情识别取得更好的性能。4.针对微表情的特点,使用鲁棒主成份分析从微表情视频中进一步的抽取细微的微表情运动信息,去除去身份信息。身份信息在微表情视频中占有很大的比重,相对与微表情识别任务来说,身份信息属于噪声。
Micro-expression is a fast leaked facial expression which is characterized by its short duration and low intensity. It can be effectively applied in lie detection as well as many other fields of studies. The research employs computer vision techniques and the research methods from cognitive psychology to develop micro-expression automatic recognition algorithms and models. Constructing two databases for micro-expression recognition. Analyzing the color space and utilize color information to further increase the accuracy of micro-expression recognition. To address the characteristic of micro-expression, we investigate the sparse representation of micro-expressions and represent micro-expressions as tensors to preserve its temporal information. Solutions presented in this report can be summarized as follows: 1. We use the psychological methods to elicit micro-expressions and use high-speed camera to capture them. Then two micro-expression databases are built and released. 2. We extend DTSA to high-order tensor and use Extreme Learning Machine to classify micro-expression.3. We analyze color space and propose a novel color space, tensor independent color space(TICS). In TICS, micro-expression recognition gets better performance. 4. For the characteristics of micro-expressions, we use Robust PCA to extract subtle motion information from micro-expression video clips.
Micro-expression is a fast leaked facial expression which is characterized by its short duration and low intensity. It can be effectively applied in lie detection as well as many other fields of studies. The research employs computer vision techniques and the research methods from cognitive psychology to develop micro-expression automatic recognition algorithms and models. Constructing two databases for micro-expression recognition. Analyzing the color space and utilize color information to further increase the accuracy of micro-expression recognition. To address the characteristic of micro-expression, we investigate the sparse representation of micro-expressions and represent micro-expressions as tensors to preserve its temporal information. Solutions presented in this report can be summarized as follows: 1. We use the psychological methods to elicit micro-expressions and use high-speed camera to capture them. Then two micro-expression databases are built and released. 2. We extend DTSA to high-order tensor and use Extreme Learning Machine to classify micro-expression.3. We analyze color space and propose a novel color space, tensor independent color space(TICS). In TICS, micro-expression recognition gets better performance. 4. For the characteristics of micro-expressions, we use Robust PCA to extract subtle motion information from micro-expression video clips.
引文