Monogenic Riesz wavelet representation for micro-expression recognition

摘要: A monogenic signal is a two-dimensional analytical that provides the local information of magnitude, phase, and orientation. While it has been applied on field face expression recognition [1], [2], [3], there are no known usages for subtle facial micro-expressions. In this paper, we propose feature representation method which succinctly captures these three low-level components at multiple scales. Riesz wavelet transform employed to obtain multi-scale wavelets, formulated by quaternion representation. Instead summing up representations, consider all representations across scales as individual features. For classification, two schemes were integrate representations: fusion-based combines features efficiently discriminately using ultra-fast, optimized Multiple Kernel Learning (UFO-MKL) algorithm; concatenation-based where combined into single vector classified linear SVM. Experiments carried out recent spontaneous micro-expression database demonstrated capability proposed in outperforming state-of-the-art approach solving problem.