针对基于仿射包的图像集人脸识别方法(AHISD)对于异常值数据的敏感性,提出了一种鲁棒性更强的方法(R1-AHISD).以仿射包模型对图像集建模,通过R1-PCA算法获得仿射子空间的正交基,进而计算定义的仿射包之间的距离,以最近邻分类器得到分类结果.在Honda/UCSD数据库上的仿真实验表明,本方法可以有效地提高识别率和对异常值数据的鲁棒性.
Since affine hull based face recognition from image sets(AHISD) is sensitive to outliers,a robust method is proposed called R1-AHISD. First, model every image set as affine hull, estimate affine subspace with rotational invariant L1 norm principal component analysis (R1-PCA), distance between hulls is defined and computed, nearest distance classifier is used to obtain the final recognition result. Experiments on Honda/UCSD face datasets show that our proposed method obtain higher recognition rates and robustness to outliers.