AUTHOR=Gao Junli , Chen Huajun , Zhang Xiaohua , Guo Jing , Liang Wenyu 

TITLE=A New Feature Extraction and Recognition Method for Microexpression Based on Local Non-negative Matrix Factorization

JOURNAL=Frontiers in Neurorobotics

VOLUME=Volume 14 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2020.579338

DOI=10.3389/fnbot.2020.579338

ISSN=1662-5218

ABSTRACT=Micro-expression has the characteristics of short duration and small action range. The existing expression recognition algorithms are not suitable for micro-expression recognition. As a feature extraction method, NMF (nonnegative matrix factorization) can decompose the original data into different components, which has been successfully applied to facial macro-expression recognition. In this paper, local NMF (LNMF) is used to decompose micro-expression into some facial muscle action, extract features for micro-expression recognition. The micro-expression datasets have not enough samples to train a classifier with good generalization performance and the existing macro-to-micro (MtM) algorithm can’t meet nonnegative properties of NMF feature vectors. As for these problems, we propose an algorithm to generate new micro-expression samples by means of macro-expression samples based on LNMF. Our algorithm has a higher recognition rate compared with three related algorithms based on CK+48/CASME2 datasets.