Facial expression recognition using expression-specific local binary patterns and layer denoising mechanism

In this paper, a novel framework for facial expression recognition is proposed, which improves the conventional feature extraction technique to further exploit distinctive characters for each label. To reduce the effect from unrelated features for facial expression recognition, a denoising mechanism is introduced. After denoising, to keep the connection between expression labels and whiten features as well as reduce the amount of computation, a manifold learning algorithm is applied, which finding a meaningful low-dimensional structure hidden in the whiten feature space. Finally, the features in the low-dimensional space are fed into the well know classifier such as the support vector machine and k-Nearest Neighbors. Simulations show that the proposed framework achieves the best recognition performance against existing methods in facial expression recognition.