Face recognition based on constructive neural networks covering learning algorithm
Guohong Huang, Zhihua Xiong, Huihe Shao · 2005
A general and efficient design approach using covering neural classifier to cope with the high-dimensional and small sample size problem is proposed. For alleviating the computational burden, face features are first extracted by the principal component analysis (PCA). In order to avoid the influence of outlier classes and reduce the large overlapping of neighboring classes, a new weighted Fisher linear discriminant (WFLD) criterion is presented by weighting contributions of individual class pairs according to the Euclidian distance of the respective class means. A new learning algorithm is used to train the neural networks classifier, which uses the "sphere neighborhoods" to cover the input samples and draw up their distributions in the original space. Thus, the training problem of neural networks may be transformed into the covering problem of a point set, which avoids the iterative process, and overcomes the problem of longtime training of classical neural networks. Simulation results conducted on the ORL database show that the system achieves excellent performance both in terms of error rates of classification and learning efficiency.