<![CDATA[Microexpression is an emotional representation occurring spontaneously and cannot be controlled consciously. It is temporary (short duration) with subtle movements, making it difficult to detect with the naked eye. Microexpressions’ muscle movements are generated in only a few small areas of the face, so observation of specific areas results in faster computation time and provides important information compared to observation of the entire face. This research proposes reducing the observation area and phase for microexpression recognition. The observed areas in the Chinese Academy of Science Micro-Expressions (CASME II) dataset are left and right eyebrows, right and left eyes, and mouth. The observation phase of microexpressions included analyzing the comparison in the onset to offset phase (“fullOAO”) and in the onset, apex, and offset phase (“OAO”). Feature extraction was performed using a simple local binary patterns histogram (LBPH) method, which can represent local features in the facial area. The best result of the proposed method was the “fullOAO” phase with an accuracy of 96.8% (using support vector machine-radial basis function, SVM-RBF) and an average computation time of 0.192 ms per frame and 10.473 ms per video. In “OAO” phase type, an accuracy of 87.7% was achieved with a computation time of 0.159 ms per frame and 0.576 ms per video. The difference in accuracy and computation time between the two-phase types occurs because the number of frames in “fullOAO” type is greater than in “OAO”, resulting in a different amount of processing time and feature extraction data. However, the 9% decrease in accuracy does not significantly affect the accuracy since the accuracy rate is still relatively good, above 80%. Furthermore, the correct measurement for computation time was the time taken to process each frame in the input video. Therefore, the proposed method can produce fast computation time and relatively accurate recognition.]]>
