Face recognition

Abstract

Much research has been done in the field of biometric applications. A computer or a system that recognises a face could the applied in various useful applications. Such as, in law enforcement and surveillance, information security, smart cards and entertainment. The scope of this thesis is to model two recognition systems and compare the performance between them. The Eigenface Approach system being a traditional system but still in use today was implemented. The basic idea behind the Eigenface Approach is to apply the Principal Component Analysis (PCA) to seek a number of eigenvectors that best describe the training database. The other face recognition system that was implemented uses Discrete Cosine Transform (DCT) coefficients amplitudes in the feature selection stage. A total of three selections tests were examined: the first one based on the average of the coefficients' amplitude; the second one involving the counting of occurrence of each coefficients; while the third test was based on the average position of the coefficients. The classification stage of the Eigenface Approach was performed using the Euclidean distance while for the DCT method a distance measure was implemented. Recognition was then performed on the different human face images using the Olivetti Research Laboratory (ORL). Both systems were modelled using the MALAB computing language and all experiments were carried out using the leave-one-out test rule. The Eigenface Approach was applied by varying the number of eigenfaces and an average recognition rate of approximately 74% was achieved. DCT Coefficients selection approach was implemented with varying amount of coefficients and an average recognition rate larger than 90% was reported for all three methods. Further work can be done by using a face image database containing of face images with a more complex background. The systems can also be extend further by being implemented in implemented in a real time environment where a subject is first detected and then identified.