Facial Expression Recognition Based On Pixel-Pattern-Based Texture Feature (PPBTF)

Facial expression is one of the most powerful and immediate means for humans to communicate and transfer information. Presently, due to its applications on sociology and computer vision, automatic facial expression recognition has attracted more and more attention.Feature extraction plays an important role, and is one of the most pivotal steps in expression recognition system. Principal component analysis (PCA), independent component analysis (ICA), Gabor wavelets and Local Binary Pattern (LBP) have been widely used for representing facial feature. Motivated by them, in this paper, a new facial representation is proposed. PPBTF (pixel-pattern-based texture feature) is a kind of texture feature, whose performance is investigated adequately in this paper. PPBTF is used for representing facial feature, and widely experiments demonstrate that it is an effective facial representation.Using Adaboost and Support Vector Machine (SVMs), the method based on PPBTF for facial expression recognition is proposed, and a real-time expression recognition system is designed. Adaboost is used for selecting the most discriminant feature subset and SVM is adopted to classify facial expression. The system can recognize seven general expressions (happy, angry, sad, disgust, surprised, fear, neutral). It is free of influence of illumination and is speedier in computation.Based on Chon-Kanade Database, JAFFE Database and many pictures from the World Wide Web (WWW), extensive experiments illustrate that the PPBTF is effective and efficient for facial expression recognition. For classification alternatively, the best recognition result achieves 100% and the average rate is also over 92%. Experiments on PIE illumination database and the authors' photos validate that the proposed method based on PPBTF is robust to illumination changes.In the last section, using the same classification, and based on the same training samples and testing samples, the PPBTF is compared with Gabor wavelets. The comparison demonstrates that the PPBTF is effective and efficient, and the facial expression recognition based on PPBTF is excellent.