DIMENSIONALITY REDUCTION AND PATTERN CLASSIFICATION METHODS FOR CRIMINAL SUSPECT FACE RECOGNITION
Ning Hui-jun · Journal of Shaanxi University of Science & Technology · 2008
This paper compares the most popular dimensionality reduction and pattern classification methods used in criminal suspect face recognition.Two subspace methods,Principle Component Analysis(PCA) and Linear Discriminate Analysis(LDA),are introduced for dimension reduction.Nearest Neighbors(NN) and Support Vector Machine(SVM) are adopted as the pattern classification methods.Experiments are conducted on the ORL face database,and the performance of different methods are presented,analyzed and compared.Experiment results show that PCA performs as well as Enhanced Fisher Model,while Enhanced Fisher Model performs better than original Linear Discriminate Analysis.And it is also found that SVM performs better than NN while SVMs with different kernels perform equally well.