Quotient vs. Difference: Comparison Between the Two Discriminant Criteria in 2DLDA for Face Recognition

摘要

Feature extraction based on two dimension image matrices projection technique are extensively employed in many face recognition methods. It does not need to transform image matrix into a vector prior for feature extraction so that it is computationally more efficient and can avoid the small sample size problem. Two discriminant criteria - quotient and difference, are commonly used in linear discriminant analysis. The performances of the two dimension linear discriminant analysis based on the two criterion are compared by face recognition experiments on two face databases. The results show that Quotient-2DLDA is usually better than Difference-2DLDA. Through theoretical analysis, the defect of Difference-2DLDA, the correlation among feature vectors is revealed, while they are uncorrelated mutually in Quotient-2DLDA. From this view point the quotient criterion 2DLDA is superior to the difference criterion 2DLDA in general.