Modified CNN-LSTM for Pain Facial Expressions Recognition

Abstract

Computer vision studies provide efficient methods for facial expressions recognition. As soon as deep machine learning approaches became involved in producing an automatic facial expressions recognition revealed even enhanced performance. In spite of the improvements have already existed in this field, researches are still need enhancements toward pain expressions recognition. To deal with the gap in existing researches, the proposed method is a hybrid model consists of a combination of the Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) model. The proposed method contribution is to successfully detect and classify the pain expression from a video sequence. The proposed hybrid model utilizes three significant factors: the spatial data presented for each frame, temporal data for each pain expression pattern, and variant face resolution. The proposed hybrid combination examined on the widely present UNBC-McMaster Shoulder Pain dataset. The results of the proposed process success to detect the pain expressions for the video sequence with an accuracy of 73.31%.