Facial expression recognition and its application based on curvelet transform and PSO-SVM

A novel method is proposed for facial expression recognition combined curvelet transform with improved support vector machine (SVM) based on particle swarm optimization (PSO). The whole process is as follows. Firstly, as wavelet transform in two-dimension is good at isolating the discontinuities at edge points and only captures limited directional information, the curvelet transform is applied to extract facial expression feature substitutively. However, the amount of curvelet coefficients obtained in the first stage is too huge to be classified, therefore, all of the coefficients are sorted descendantly and the former larger 5 or 10% are remained while the others abandoned to reduce the dimension. Finally, PSO algorithm is employed to search for the reasonable parameters of SVM to increase classification accuracy. Experimental results demonstrate that our proposed method can form effective and reasonable facial expression feature, and achieve good recognition accuracy and robustness, which is competent for spirit states detection of operators to decrease defect rate of production.