Face location by template matching with a quadratic discriminant function

Face Detection: A Survey

“…Instead of computing a transformation matrix (the [u (20)) aimed for representation of the entire dataset (PCA), LDA seeks to find a transformation matrix which is based on maximizing between class scatter and minimizing within class scatter. The eigenvector in the LDA transformation matrix with the largest eigenvalue is known as Fisher's linear discriminant [43], which by itself has also been used for face detection [179,203]. PCA is aimed at representation, while LDA aims for discrimination and is therefore appropriate for face detection when the class of faces and nonfaces is divided into subclasses.…”

Face verification for mobile personal device

“…LDA is a supervised dimensionality reduction approach, seeking to find a projection matrix which maximally discriminates different classes [52]. Generally speaking, LDA is intended for a multi-class problem, but it can also be applied to the two-class face detection problem when the class of face and nonface are clustered into subclasses [183] [154]. This allows more complicated modeling of the face space.…”