Face Detection Based on LDA and NN
Hiroyuki Kobayashi, Qiangfu Zhao · 2007
In this paper, we propose a new method for face detec- tion by combining a modified linear discriminant analysis (M-LDA) and neural network (NN). Here, M-LDA is used for feature extraction, and NN is used to make the final deci- sion. The M-LDA minimizes the variance within all "face" clusters, and at the same time, maximizes the variance be- tween all "face" clusters and all "non-face" patterns. The feature space obtained by M-LDA has a dimensionality less than that of the original problem, and thus the complexity of the NN used in the second step can be greatly reduced. To validate the efficacy of the proposed method, we con- ducted several experiments with four methods, namely the proposed method, NN, PCA+NN, and LDA+NN. Results show that the proposed method can provide lower false pos- itive and false negative errors for unknown test images.