Low-dimension Presentation of Facial Pose and Expression
Kun He · Jisuanji gongcheng · 2009
The variety of facial pose and expression will affect the result of facial recognition.The traditional methods analyses facial pose and expression from every pixel of image on the contrary.This paper analyses the pose and expression from geometric relationship among image data.It uses Locally Linear Embedding(LLE) which computes low dimensional,neighborhood preserving embedding of high-dimensional data.The algorithm finds the nonlinear structure of data effectively,invariable to translation,revolution and so on.Experiment shows that the algorithm has capability to reflect the slight variety of facial pose and expression,and indicates that the low-dimensional distribution of various pose and expression is identical to different persons and the original image can be reconstructed exactly from neighboring images.