Discriminative feature extraction and selection applied to face recognition

摘要: We propose an integrated approach to feature and architecture optimization for convolutional connectionist models. The goal is select single features which are likely have good discriminatory power extract nonlinear combinations of with the same aim. In particular, focus on interaction extraction selection modules recognizer design. a pruning-based method called /spl epsi/HVS (extended HVS), where use priori knowledge adaptively optimized during discrimination training criterion aiming at minimum classification error. Results demonstrate approach's effectiveness in identifying reduced architectures recognition accuracy.