Recognizing Facial Actions Using Gabor Wavelets with Neutral Face Average Difference
Juliano J. Bazzo, Marcus V. Lamar, Rui Seara

DOI: 10.14209/sbrt.2004.68
Evento: XXI Simpósio Brasileiro de Telecomunicações (SBrT2004)
Keywords: Face recognition facial action units Gabor wavelets artificial neural network
This paper describes a new pre-processing strategy to classify facial expression. Previous research works suggest that Gabor wavelets applied to recognize facial expression images subtracted from the neutral face of the same subject could provide a satisfactory recognition rate under controlled conditions, such as eye and mouth alignments. Here, we propose a recognition system in which the Gabor kernels are applied to a facial expression subtracted from an averaged neutral face. A pre-processing stage which generates a reduced output data set is also proposed. By using an artificial neural network-based classifier, recognition rates of 86.6% and 81.6% are obtained for 7 upper and 7 lower face actions, respectively. Considering a heterogeneous subject database with head motion and lighting variations, the new recognition approach is assessed for performance. The obtained results attest the effectiveness and applicability of the proposed technique for recognizing facial actions.