With large amounts of patches randomly selected from training faces, the model was trained firstly via backpropagation which minimizes an unsupervised sparse reconstruction error, and then a softmax classifier was learned for supervised classification. 6 gave better results. We have decided to convert the outputs into a binary format. Vitria, online nonparametric discriminant analysis for incremental subspace learning and recognition, pattern anal applic, vol.
3d face recognition system matlab code
Facial expression code. That all of these five cases gave an additional classification in addition to the correct classification (e. Based on initial testing, the best performing generalized and specialized neural networks were recruited into decision making committees which formed an integrated committee neural network system. It scores between 91% and 100%, depending on which emotion is measured, when comparing facereader outcomes with the facial expressions scored manually by the professional annotators – van der schalk, j. Shows the average percentage presence of the seven binary parameters for all of the seven expressions. Real-valued measures were obtained for expressions including the neutral image. Final evaluation of the committee is conducted with data obtained form subjects not used in training or in initial testing.
These four expressions are negative emotions which are often difficult to classify. The question remains if a committee or committees of neural networks trained on back-propagation can provide a reasonable (close to 90%) accuracy in classification of different facial expressions.