Training of Feature Extractor via New Cluster Validities for Adaptive Facial Expression Recognition
Sang Wan Lee, Dae‐Jin Kim, Yang Soo Kim, Zeungnam Bien · 2005
A lot of researches on classifier, which can perform well with a given set of feature vectors, have been done. However, researches on feature vectors, that extract better feature vectors automatically, have not been done very much, not to mention of considering both issues at the same time. Regarding classification, 'adaptation' becomes very important issue, because person's way of making facial expression is changed as time goes by. On the other side, we face two problems when we consider the issue, 'feature extraction'. One is how we can make a good feature extractor, and the other is what more separable features are. Proposed method solved these two problems by proposing adaptive Gabor Wavelet Neural Network for facial expression recognition which utilizes novel cluster validity measures as an objective function. In training stage, we use an expectation value of all cluster validities combined with variance in such a way that they tend to be minimized at the same time. The simulation results show overall recognition rate of 97%.