Subpixel subtle motion estimation of micro-expressions multiclass classification
Niyalatul Muna, Ulla Delfana Rosiani, Eko Mulyanto Yuniamo, Mauridhi Hery Purnomo · 2017
Communication between one individual and another can be delivered in different ways, not only through speech, writing and body motions. Communication can also be delivered through basic emotions one feels through visible facial expressions. Facial expressions displayed in the forms of emotional expressions can be demonstrated briefly or quickly, which is known as micro-expression. The subtle motion of micro facial expression in every exchange makes many people find it difficult to identify and recognize the ongoing emotion. Therefore, this study proposes the recognition of micro-expression emotions with estimated subtle motion in the image sequence based on onset-frame, apex-frame, and offset-frame for feature extraction. The method used for feature extraction are the combination of Block Matching Algorithm with Taylor Series Approximation referred to as Subpixel Subtle Motion Estimation (SME). Multiclass Classification process using Multilayer Perceptron (MLP) Backpropagation and Support Vector Machine (SVM) as a comparison. The evaluation results show the best accuracy 85.07%, with the Mean Absolute Error 0.0597 and Root Mean Square Error 0.2443.