Facial Feature Tracking, Extraction and Selection

In this report, first we explain how we moved toward implementing an algorithm for tracking facial features. The explanations, thoroughly covers our experiences, both our failures and successful trials. As a failed experienced of facial feature tracking, we have explained the correlation-based tracking and its improved extension. The second tracking method, is devoted to Active Appearance Model and its details. The implementation of the explained model showed that the AAM model has enough ability for real-time facial feature tracking. Simultaneously, we followed extraction of facial features using Gabor wavelet and selecting facial feature points algorithms. Eventually, due to high dimension of extracted feature vectors derived by the use of Gabor wavelet, we moved to selecting the most elite sup-population of the features which will result in the most discriminative classification procedure and have lowest redundant information. Hence, Adaboost method and PCA based algorithms are exploited so as to reduce the dimension of the generated feature vector. The comprehensive explanation of the feature selection methods has been brought in the main context.