Mastering OpenCV 3 - Second Edition

Linear shape models

The aim of facial-deformation modeling is to find a compact parametric representation of how the face's shape varies across identities and between expressions. There are many ways of achieving this goal with various levels of complexity. The simplest of these is to use a linear representation of facial geometry. Despite its simplicity, it has been shown to accurately capture the space of facial deformations, particularly when the faces in the dataset are largely in a frontal pose. It also has the advantage that inferring the parameters of its representation is an extremely simple and cheap operation, in contrast to its nonlinear counterparts. This plays an important role when deploying it to constrain the search procedure ...