Abstract:

Aiming at the problem of cross-database micro-expression recognition, a micro-expression recognition method based on Apex frame optical flow and convolutional autoencoder is proposed. The method includes three parts：preprocessing, feature extraction and micro-expression classification. The preprocessing section performs Apex frame positioning, face detection and alignment on the micro-expressions. The feature extraction section first calculates the TVL1 optical flow of the pre-processed Apex frame, then uses the obtained horizontal and vertical optical flow component images to train the convolutional autoencoder to obtain the optimal structure and parameters, finally combines the two components from the features of the middle layer of the encoder as the features of the micro-expressions. In section of micro-expression classification, a Support Vector Machine（SVM） classifier is used to classify the features extracted in the previous step. The experimental results have been greatly improved compared to the baseline method（LBP-TOP）. Among them, UF1 has increased by 0.134 4, and UAR has increased by 0.140 6. This method provides new ideas for micro-expression features extraction and recognition.

Key words: micro-expression recognition, Apex frame, optical flow, convolutional autoencoder, Support Vector Machine（SVM）

摘要：

针对跨库微表情识别问题，提出了一种基于Apex帧光流和卷积自编码器的微表情识别方法。该方法包括预处理、特征提取、微表情分类三部分。预处理部分对微表情进行Apex帧定位以及人脸检测和对齐；特征提取部分首先计算预处理过的Apex帧的TVL1光流，然后使用得到的水平和竖直光流分量图像训练卷积自编码器得到最优结构和参数；最后将两个分量自编码器中间层的特征融合后作为微表情的特征；微表情分类就是使用支持向量机（Support Vector Machine，SVM）对上一步中提取到的特征进行分类。实验结果较基准方法（LBP-TOP）有了很大的提高，UF1提高了0.134 4，UAR提高了0.140 6。该方法为微表情特征提取和识别提供了新的思路。

关键词: 微表情识别, Apex帧, 光流, 卷积自编码器, 支持向量机（SVM）