Automatic facial feature extraction and facial expression recognition

In this paper, we present an automatic algorithm for facial expression recognition. We first propose a method for automatic facial feature extraction, based on the analysis of outputs of local Gabor filters. Such analysis is done using a spatial adaptive triangulation of the magnitude of the filtered images. Then, we propose a classification procedure for facial expression recognition, considering the internal part of registered still faces. Principal Component Analysis allows to represent faces in a low-dimensional space, defined by basis functions that are adapted to training sets of facial expressions. We show how to select the best basis functions for facial expression recognition, providing a good linear discrimination: results prove the robustness of the recognition method.