Ameliorate grasshopper optimization algorithm based long short term memory classification for face emotion recognition system

Face Emotion Recognition (FER) has become a vital need for human interaction with machines in the virtual world. In virtual classroom, the emotional states of the students are necessary for better learning and communication. In this work, Ameliorate Grasshopper Optimization Algorithm (AGOA) based Long Short Term Memory (LSTM) classification for Face emotion recognition system is presented. AGOA is proposed by enhancing the conventional Grasshopper Optimization Algorithm (GOA) with opposition-based learning, levy flight mechanism and Gaussian mutation for the selection of optimized features. Convolution Neural Network (CNN) based feature extraction is adopted and the LSTM unit classifies the basic human emotions such as fear, happy, disgust, anger, sad, surprise and normal. The exploratory work is done using the YALE face database and it resulted in 93.90% recognition accuracy. The results attained shows that the performance measures such as precision, recall, specificity, F-Measure, sensitivity and Area under Curve (AUC) are higher in AGOA-LSTM based system rather than the system without AGOA and it also resulted in reduced error rate.

Automated summary

This work proposes a face emotion recognition system based on AGOA-LSTM, which utilizes CNN for feature extraction and classifies seven basic human emotions. Experimental results show that the system achieves high recognition accuracy on the YALE face database, with significantly better performance measures compared to the system without AGOA.