Combinding KFLD-ISOMAP and SVM for face recognition
Junying Gan, Yong-Hui Kuang · 2011
Dimensionality reduction has always been a key problem in the fields of face recognition. In this paper, we present an effective method for face recognition, in which we use Isometric Mapping (ISOMAP) for estimating geodesic distance between data points and then apply Kernel Fisher Linear Discriminant (KFLD) to find the projection that maximizes the distances between cluster centers and also could deal with the data that can not be linearly separated in the low-dimension. By means of KFLD-ISOMAP, the optimal feature vectors of the samples perform well using Support Vector Machine (SVM) classifier in face recognition. Experimental results on Olivetti Research Laboratory (ORL) and Yale face database demonstrate that the method of KFLD-ISOMAP and SVM is more effective and robust than the original ISOMAP, FLD-ISOMAP and some traditional methods in face recognition.