Statistical Approaches To Face Recognition And Performance Index

The purpose of this thesis is to study statistical approaches to face recognition(FR) and the corresponding performance evaluation methods. Firstly, some mainstream methods of FR are introduced briefly, and the system design is presented as well as some important performance indexes. Secondly, some feature extraction methods based on statistics are studied in detail, including Principle Component Analysis(PCA),Linear Discriminant Analysis(LDA) and Independent Component Analysis(ICA) etc. The problem of selecting the dimensions of projection space and feature classification is analyzed. With plenty of experiments, some novel viewpoints are proposed about dimension selection and distance measure. In addition, performance indexes are discussed, the way to calculate them and the relations between them are explained, which is helpful for system design. Finally, base on above decisions, a whole design layout is presented and a demo system is realized on VC++ IDE environment. Experimental results show that the demo system does work.