论文-3D人脸对齐方法2

Towards Pose Robust Face Recognition

CVPR2013

This paper will mainly focus on the pose problem whileconsidering the other factors together

这方面已有的工作（tackle the problem of recognizing faces in arbitrary poses）：

1.    Illumination Cone Model (ICM)

2.   EigenLight Field (ELF)

3.   3D Morphable Model (3DMM)

已有工作可分为两大类：

1.   2D methods（2D methods often use some 2D transformations）

piecewise affine,

thin plate splines

2.  3D methods(flexibility\ precision)

a single model

a deformable model

3D methods can be dividedinto four categories depending on how to use of the 3D model:

本文致力于第四种方法：filter transformation

the main idea of which is transformingfilter according to the pose and shape of face image and then using thetransformed filter to extract pose robust features. As the filter varies withthe change of pose, so we call the method as “Pose Adaptive Filter” (PAF).

本文框架：

Firstly, a 3D deformable model is built and a fast 3D model fittingalgorithm is proposed to estimate the pose of face image.

Secondly, a group of Gabor filters are transformed according to thepose and shape of face image for feature extraction

Finally, PCA is applied on the pose adaptiveGabor features to remove the redundances and Cosine metric is used to evaluatethe similarity