Boosting N-Tuple Networks for Face Recognition
Hong Peng · 2005
The traditional N--tuple network consists of KRAM neurons, and each RAM is formed by N tuples that can be obtained by sampling input sample which is firstly converted to binary vector. The improved version of N--tuple network is obtained by replacing RAM neurons with sparse ones, so that real input can be dealt with directly there. On the other hand, it is the fact that network scale is out of control due to elusive experiential value N and K. We apply Boosting algorithm to the improved N--tuple network. Finally, some results are obtained as follows: (1) Different fundamental classifiers perform on different features for identical training sample as well as each base classifier performs on different training sets; (2) It will be shown that our proposed method implies Boosting can be applied to high dimension data directly; (3) The improved performance by Boosting is almost independent from the scale of network; (4) When this method is applied to face recognition, no feature extraction and selection procedures are obviously performed.