Design And Implementation Of Micro-expression Recognition System

Micro-expressions are powerful in conveying useful,latent information.However,it is infeasible for them to be captured by the human eye,due to their short duration and obscureness.Hence,it is necessary and important to develop a system to detect microexpressions automatically.In this project,a micro-expression system is developed based on machine learning in MATLAB,including research on background information,architecture design,implementation,training,testing and evaluation.The system adopts local binary pattern on three orthogonal planes(LBP-TOP)for feature extraction and back-propagation neural network(BPNN)is utilised to train detection models capable of fetching scenes with micro-expressions from video clips.Support vector machine(SVM)is then implemented to classify five micro-expressions(happiness,repression,surprise,disgust and others).The models are trained by varying different parameters and are evaluated against a CASEME2 database.The models yielding the best performance are then used to detect and classify micro—expressions from videos.According to the experimental results,the detection model is acceptable and the classification model is relatively strong.Finally the Micro-expressions Recognition(MER)system was successfully developed.The system consists of two parts,including detection of,and classification when retrieved micro-expressions are labelled.The results suggests that many factors could cause huge influences on performances of the system,such as choice of system architecture and parameters.