Publication Date

Spring 2018

Degree Type

Master's Project

Degree Name

Master of Science (MS)


Computer Science


Micro-expressions are short-lived, rapid facial expressions that are exhibited by individuals when they are in high stakes situations. Studying these micro-expressions is important as these cannot be modified by an individual and hence offer us a peek into what the individual is actually feeling and thinking as opposed to what he/she is trying to portray. The spotting and recognition of micro-expressions has applications in the fields of criminal investigation, psychotherapy, education etc. However due to micro-expressions’ short-lived and rapid nature; spotting, recognizing and classifying them is a major challenge. In this paper, we design a hybrid approach for spotting and recognizing micro-expressions by utilizing motion magnification using Eulerian Video Magnification and Spatiotemporal Texture Map (STTM). The validation of this approach was done on the spontaneous micro-expression dataset, CASMEII in comparison with the baseline. This approach achieved an accuracy of 80% viz. an increase by 5% as compared to the existing baseline by utilizing 10-fold cross validation using Support Vector Machines (SVM) with a linear kernel.