Automated Facial Micro-expression Recognition Using Local Binary Patterns on Three Orthogonal Planes with Boosted Classifiers: A Survey

Abstract

Micro-expressions are used to understand one’s mind and also show hidden intentions through facial subtle changes. SwanCare is using facial recognition technology to identify micro-expressions that indicate the presence of pain in elderly patients with dementia who cannot verbally express the pain. Recognition of such micro-expressions needs trained experts as evidenced by lie detection and polygraph experts. Automatic expression recognition not only removes subjectivity but also saves resources and time. The paper proposes the detection of micro-expressions using the local binary patterns on three orthogonal planes(LBP-TOP) feature extractor due to its capability to detect temporal facial expression features. Other variants of the algorithm, namely LBP-MOP which is a mean of the feature extractions and LBP-SIP were also analyzed. The weighted classifiers proposed included support vector machines, k-nearest neighbor and random forest as a weighted ensemble voting classifier. To evaluate the accuracy and performance, the CASME II and SMIC 3D datasets were used in the study. A higher accuracy was obtained which was an improvement to other modern micro-expression methods. The study also leveraged Gabor wavelet filters, principal components analysis (PCA) and linear discriminant analysis (LDA) for dimension reduction.