Facial Expressions Recognition Based on Cognition and Mapped Binary Patterns

In this work, we extract the facial contours by seeking the largest island method based on Local Binary Pattern (LBP) and segment the facial regions by establishing pseudo- 3D model which makes the gray values of the images as its z-axis. Moreover, we use Multi-Dimensional Scaling (MDS) and the approximation of Earth Mover's Distance (EMD) to reduce the dimension of data. 

Once more, we use the feature images reduced dimension to train the Convolutional Neural Network model to predict the expressions and also choose the different feature regions or combinations as the reference to find out the best distinguished region or combination and compare with the conclusions in cognitive neurology. Finally, we have a comparative experiment between traditional basic emotion model and dimension space model.