Design of ResNet50-LSTM Model with Attention Mechanism and Transfer Learning for Microexpression Recognition

Our face can show different emotions. From the feelings, one can identify the emotional state of a person. Micro expression is a subtle facial movement that reveals a person's subconscious feelings. Detecting micro-expressions is essential to detect the human emotional state, which is very helpful for the therapist or psychologist. This paper presents a method of micro expression recognition using the model that combines ResNet50-LSTM with an attention mechanism and Transfer Learning. ResNet50 captures the spatial features, LSTM for temporal dependencies between frames. The attention mechanism finetuned the built model. We adopted transfer learning to enhance the model's performance in detecting subtle micro-expressions. Experiments were conducted on benchmark micro-expression datasets such as SAMM and CASME II. The results show the model's superior performance in recognising emotions, which can help understand human psychology.