I am trying to understand Eigenfaces Vs. Fisherfaces: Recognition Using Class Specific Linear Projection paper. It uses PCA and further uses LDA for dimensionality reduction.

I have read about eigenfaces and PCA approach but I am not familiar with LDA and I am unable to understand the mathematics behind the scatter matrices : why to minimize the determinant of projected scatter matrices and how we perform it.

Can anyone explain the maths behind LDA and how the generalized eigenvalues come into picture or point to some related tutorial where I can understand the approach?

  • 1
    $\begingroup$ Not specifically in relation to eigenfaces, LDA theory and math (for k class situation) have been covered several times on this site. See especially user amoeba's and mine answers (discriminant-analysis tagged). For 2-class simpler situation, LDA has been explained by many here. $\endgroup$
    – ttnphns
    Commented Dec 6, 2017 at 12:16
  • $\begingroup$ Thanks found bit too many links!! Any good book/paper to follow the maths involved (too many links makes it a bit confusing)? $\endgroup$
    – Naman
    Commented Dec 6, 2017 at 12:34