Facial emotion recognition using subband selective multilevel stationary wavelet gradient transform and fuzzy support vector machine

摘要

Facial emotion recognition finds a major role in affective computing. Recognizing emotion by facial expression is an extremely important activity to design control oriented and human computer interactive applications especially in cognitive science and neuroscience. For a precise and robust recognition, feature extraction is one of the major challenges in facial expression recognition system. Wavelet transform is one of the major key methods utilized for feature extraction in facial emotion recognition. In this paper, the statistical parameters from the proposed subband selective multilevel stationary wavelet gradient transform are calculated and are utilized as features for efficacious recognition of emotion. The features of the wavelet transform contain both spatial and spectral domain information which is best suited for identifying human emotions through facial expression. The introduction of gradient transform to find the gradient of subband avails to estimate the edges in images for the quality amelioration of subbands. The dimension reduction in the extracted features is done by using Pearson-kernel-principal component analysis method. The classification of emotion using the selected features is done by the proposed Gaussian membership function fuzzy SVM classifier. Experiments were performed on the well-known database for facial expression such as JAFEE database, CK + database and FG Net database and obtained promising emotion classification results.