针对人脸识别中小样本问题导致类依赖子空间不完善而严重影响识别性能的问题，提出一种基于线性判别回归的最近-最远子空间分类算法。首先，基于线性判别回归，利用最近子空间分类器度量测试图像与单一类之间的关系；然后，利用所提出的最远子空间分类器度量测试图像与训练图像之间的关系；最后，结合最近、最远子空间分类器，利用类依赖子空间的不同特性完成人脸的分类识别。在三个公开的人脸数据库ORL、AR及扩展YaleB上的实验验证了该算法的有效性。实验结果表明，相比其他几种分类算法，该算法取得了更好的识别效果。%The small sample problem in face recognition causes the imperfection of class dependent subspace and the recognition performance is seriously impacted,in light of this problem,we propose a LDR-based nearest-farthest subspace classification algorithm.First, it uses nearest subspace classifier to measure the relation between testing images and single class based on linear discriminative regression. Then,it uses farthest subspace classifier to measure the relation between testing and training images.Finally,it combines the nearest and farthest classifiers to complete face recognition by using different characters of class dependent subspace.The effectiveness of the proposed algorithm has been verified by the experiments on three common databases ORL,AR and extended YaleB.Experimental results show that the proposed algorithm achieves better recognition effect than several other classification algorithms.
展开▼
