Dynamic Facial Expression Understanding Using Deep Spatiotemporal LDSP On Spark

摘要: Facial expressions are the most common medium for expressing human emotions. Due to wide range of real-world applications, facial expression understanding has received extensive attention from researchers. One vital issues recognition is extraction and modeling temporal dynamics emotions videos. Additionally, rapid growth video data various multimedia sources becoming a serious concern. Therefore, address these issues, in this paper, we introduce novel approach on top Spark First, propose new dynamic feature descriptor, namely, local directional structural pattern three orthogonal planes (LDSP-TOP), which analyzes aspects texture. Second, design 1-D convolutional neural network (CNN) capture additional discriminative features. Third, long short-term memory (LSTM) autoencoder employed learn spatiotemporal Finally, an experimental investigation carried out demonstrate performance scalability proposed framework.