Weighted Two-Dimensional Heterosecedastic Discriminant Analysis for face recognition

In Two-Dimensional Linear Discriminant Analysis (2DLDA), it is satisfied that within-class covariance matrixes are equal; while in Two-Dimensional Heteroscedastic Discriminant Analysis (2DHDA), within-class covariance matrixes are heteroscedastic. Based on the characters of 2DLDA and 2DHDA, Weighted Two-Dimensional Heteroscedastic Discriminant Analysis (W2DHDA) is introduced and used in face recognition, in which within-class covariance matrix is defined as weighted summation of both within-class covariance matrixes of 2DLDA and 2DHDA. In this way, the defined within-class covariance matrixes in W2DHDA are more robust. Experimental results based on ORL (Olivetti Research Laboratory) and Yale mixture face database show the validity of W2DHDA in face recognition.