Robust Face Recognition Based on Modified ICA without Training Sample of Test Subjects

At present there are many methods that could deal well with frontal view face recognition when there is sufficient number of representative training samples. There into, subspace learning method such as principal component analysis (PCA), independent component analysis (ICA), linear discriminant analysis (LDA) are a very hot research topic in this field. However, in some face recognition system, the needed recognition faces are different with different users and are often changed according users´ requirement. So in this paper, we proposed to make use of some known face database, in which subjects will be different with the test subjects, for training (extract subspace) with a modified ICA method. The proposed modified ICA method can save much of computer time and memory, and at the same time, can obtain acceptable experimental results on a part of FERET face database. In addition, we validate that the accuracy rate with simple rotation images of logged faces as known data can be improved when the logged face is only one sample per subject.