Facial Expression Recognition

Facial expression recognition is a developing research topic in artificial intelligence field, aiming at letting artificial intelligent products (e.g. robot) recognize human facial expressions and analyze emotions automatically. Automatically facial expression rcognition plays a vital role in realizing a highly intelligent human-machine interface, so it is worthy of attracting much research and attention.Basing on the advances of current facial expression recognition technology currently, an approach for feature extraction and classification of facial expression recognition is presented in the thesis. The thesis mainly includes:1. Face detection algorithm and expression image preprocess is studied. In order to detect face and locate eye position Adaboost machine learning algorithm is adopted. Geometric preprocessing including rotation, clipping and scaling, photometric preprocessing including histogram equalization are adopted.2. Comparison of current facial expression recognition algorithm is discussed. The method of using Gabor wavelet transformation to extract features and using the transformation coefficients to substitute for the gray value is adopted, which can reduce the sensitivity to variations of lighting and position. PCA/LDA Fisher discriminant analysis is performed on Gabor vector to reduce dimension of features and train a feature subspace to maximize the between-class scatter, which can result in a higher speed of recognition and a better classifying performance.3. Generation of expression templates and classifying is studied. Based on the strategy of making use of the differences between the forms of one expression, one expression is divided into some subexpressions. An algorithm of using dynamic C-Means clustering to generate expression templates and K-Nearest neighbor to classify expression is proposed. Experiments demonstrate that this algorithm can perform a good classification on facial expression images.4. An automatic facial expression recognition prototype system is developed based on the algorithm studied in this thesis. The system gets input expression images from a camera and traces face, locates eyes in real time. At the same time it recognizes 7 expressions of happiness, fear, sadness, disgust, anger, surprise and neutral.