Abstract: An improved adaptive principal component extraction algorithm based face recognition method is proposed in this paper. Improved adaptive principal component extraction algorithm is applied to project facial images from high-dimensional observation space to low-dimensional feature space. The weight-value updating equation of feed-forward network is improved to decrease the computational complexity. Region filling and hidden-surface removal algorithms are adopted to generate multiple virtual samples according to single facial image based on 3D face morphable model which is adopted to tackle the small sample size problem. Experimental results on ORL and UMIST face database show that our method makes impressive performance improvement compared with conventional Eigenface and Fisherface methods.

Key words: Face Recognition, Adaptive Principal Component Extraction, Morphable Model

摘要： 论文提出了一种基于改进的自适应主元提取算法的人脸识别方法。采用改进的自适应主元提取算法将人脸图像由高维观测空间投影到低维特征空间，通过改进前馈网络权值更新方程，降低算法的复杂度和计算量。基于三维人脸形变模型，采用区域填充和曲面消隐算法根据一幅人脸图像生成多个虚拟样本，克服人脸识别中的小样本问题。在ORL和UMIST数据库上的实验结果表明，本文提出的算法在识别性能上明显高于传统的Eigenface和Fisherface方法。

关键词: 人脸识别, 自适应主元提取, 形变模型