Piecewise linear landmark mapping for pose normalization

This paper presents two pose normalization techniques based on landmarks and linear mappings between these landmarks. A column based and a polygon based transformation will be discussed and tested with a PCA and an LDA classifier. The results~show that the combination of the polygon transformation paired with the LDA classifier gives the best equal error rates. When using PCA, the column and polygon transformations are very close in performance. Overall, both transformations give better scores than leaving the images untouched.