基于像素特征的微表情识别

微表情持续时间短、表达强度低，给训练有效模型带来了挑战。针对此问题，提出了一种基于像素特征的微表情识别方法。对图像序列的面部区域进行裁剪，消除背景噪声；将每一帧的像素矩阵与第一帧(中性表情)做差处理，提取面部变化；对做差的结果累加，进一步突出面部表情；使用搭建的浅层CNN网络进行分类。在3个公共微表情数据集组成的交叉数据集上进行K折(K-fold)交叉验证实验中，所提方法的3个评价指标ACC(accuracy)、UF1(unweighted F1-score)和UAR(unweighted Average Recall)分别达到了0.830 4、0.782 7和0.794 4,表明了该方法的有效性。与LBP-TOP等8个模型的对比实验中，所提方法的指标明显优于对比模型，验证了该方法的优越性。

Micro-expression recognition based on pixel feature

The short duration and low intensity bring great challenges to micro-expression recognition. In response to the above problems, this paper proposes a micro-expression recognition method based on pixel feature. First, the facial area of the image sequence is cropped to eliminate background noise, and then the pixel matrix of each frame is combined with the first frame (neutral expression). Difference processing is used to extract facial changes, and then the results of the difference are accumulated to further highlight facial expressions, and finally the shallow CNN network built is used for classification. In the K-fold cross-validation on a cross dataset composed of 3 public micro-expression data sets, the three evaluation indicators of ACC (accuracy), UF1 (unweighted F1-score) and UAR (unweighted average recall) reached 0.830 4, 0.782 7 and 0.794 4, and comparative experiments with related methods such as LBP-TOP (local binary pattern on three orthogonal planes) also verified the superiority of this method.