摘要
使用人脸数字图像奇异值分解中前面部分较大的奇异值及其对应的特征向量来重构图像,以剔除原图像中由于光照、表情、姿势等噪声影响对应的高频信息,并将重构图像作为模板进行识别.此方法无需使用较多的奇异值和训练样本,就能达到了很高的识别率,大大降低了识别工作量,并优于PCA方法.
In this paper, those bigger singular values and their corresponding eigenvectors are employed as features while others are excluded as the high frequency information disturbed by the illumination, such as the expression, the pose or other noise. Experimental results suggest that this novel method works well and the recognition rates are higher than that of eigenface based on PCA.
出处
《广东教育学院学报》
2006年第3期92-96,共5页
Journal of Guangdong Education Institute
基金
教育部重点基金资助项目(104145)
广东省自然科学基金资助项目(031609)
关键词
奇异值分解
人脸识别
PCA方法
singular value decomposition
face recognition
PCA