Figure 1
From: Micro-expression recognition model based on TV-L1 optical flow method and improved ShuffleNet
Micro-expression recognition model. (1) Obtain the start frame and vertex frame from the pre-processed sequence and extract the TV-L1 optical flow features between them. (2) Feed the optical flow features into the improved ShuffleNetV2 for pre-training and get the weights. (3) Load the weights and feed the pre-processed dataset into the improved ShuffleNetV2. (4) Use the SVM classifier in training for classification. Images are from subject 1 of the CASME II dataset6 and is reproduced here following the CASME2 license agreement. Copyright © Xiaolan Fu.
