基于Overcomplete ICA的人脸特征提取

Face Feature Extraction Based on Overcomplete ICA Feature Extraction

人脸特征提取是人脸识别中最重要的一个环节,人脸特征提取的一种主要方法是寻找一系列的基图像,然后再把人脸表示为这一系列基图像的线性叠加。PCA和ICA在寻找基图像的过程中,源图像和基图像的数目都是相同的。本文提出了一种基于Overcomplete ICA的人脸特征提取方法,所得到的基图像数目要多于源图像数目。最后采用最小距离分离器进行人脸识别的实验,并与PCA和ICA的识别效果进行比较。

Face feature extraction is the key procedure in face recognition. An important approach of face extraction is to face a set of base images, then represent faces as a linear combination of those base images. During the procedure of finding base images using PCA and ICA, the number of source images is equal to base images. In the paper, we present a face feature extraction approach, which gain the the number base images is lager than source images. At the end of thts paper, we give a face recognition examination using Overcomplete ICA feture extraction and shortest dis- tant separation, and compare the result with PCA and ICA.