首页|基于滑动窗口和时空特征的微表情检测算法

基于滑动窗口和时空特征的微表情检测算法

扫码查看
微表情存在时间短、强度弱等固有特性,目前微表情自动检测仍然存在较大的困难.为了提升微表情检测的效果,论文提出一种基于滑动窗口和时空特征的微表情检测算法:首先使用滑动窗口技术将一段微表情视频分割成若干个滑动窗口,然后在每个滑动窗口中提取时空特征,并和微表情SP模式匹配得到单个窗口的检测结果,最后融合所有滑动窗口的检测结果.该算法在CAS(ME)2和SAMM数据集上进行了实验,并与2020年微表情挑战赛(MEGC 2020)的基线结果进行了对比,结果显示,在CAS(ME)2和SAMM数据集上该算法比基线算法在微表情检测上分别提升了4.7%、9.7%,在整体上分别提升了9.9%、5.7%,验证了该算法的有效性.
Micro-expression Detection Algorithm Based on Sliding Window and Spatio-temporal Features
Due to the inherent characteristics of short time and weak intensity of micro expression,it is still difficult to detect micro expression automatically.In order to improve the effect of micro expression detection,this paper proposes a micro expression detection algorithm based on sliding window and spatio-temporal features.Firstly,a micro expression video is divided into several sliding windows by using sliding window technology,then the spatio-temporal features are extracted from each sliding window,and the detection results of a single window are obtained by matching with micro expression SP pattern.Finally,the detection results of all sliding windows are fused.The algorithm is tested on CAS(ME)2 and SAMM data sets,and compared with the baseline results of the 2020 micro expression challenge(MEGC 2020).The results show that the algorithm improves the micro expression detection by 4.7%and 9.7%respectively on CAS(ME)2 and SAMM data sets,and 9.9%and 5.7%on the whole.The effectiveness of the algo-rithm is verified.

micro-expression detectionsliding windowspatio-temporal features

马崟桓、黄树成、李明星

展开 >

江苏科技大学计算机学院 镇江 212003

江苏大学京江学院电气信息工程学院 镇江 212003

微表情检测 滑动窗口 时空特征

国家自然科学基金项目

61772244

2024

计算机与数字工程
中国船舶重工集团公司第七0九研究所

计算机与数字工程

CSTPCD
影响因子:0.355
ISSN:1672-9722
年,卷(期):2024.52(6)
  • 1