一种基于对称性和上下文约束的线性鉴别分析方法

A linear discriminate analysis method based on symmetry and contextual constraints

在充分考虑人脸对称性的基础上,结合图像中像素之间的上下文约束关系,提出了一种改进的图像鉴别方法,即基于对称性和上下文约束的线性鉴别分析方法 (SCCLDA).为了证明改进算法的优势,本文进一步在原始样本和镜像样本的扩展集合上测试了CCLDA(ECCLDA)的识别性能.实验研究表明,在人脸受光照、姿态以及表情等外在因素影响情况下,SCCLDA方法比ECCLDA、CCLDA、LDA等方法在人脸识别效果上具有更好的稳定性和更高的准确性.

Linear discriminate analysis（LDA）considers the discriminative information in the process of feature extraction,but the contextual information among pixels in the high dimensional space is not exploited.Contextual constraints based linear discriminate analysis（CCLDA）incorporates the contextual information into linear discriminate analysis during feature dimensionality reduction,which can provide much more useful information for classification.In this paper,linear discriminate analysis method based on symmetry and contextual constraints is proposed.In the improved method,the symmetry of the face to generate new samples is exploited and the contextual constraint in images is considered to perform face recognition.Moreover,to show the superiority of the improved method,the recognition performance of CCLDA is tested on the extended set of the original samples and the image samples.Experiments are conducted to prove the effectiveness of SCCLDA by varying illumination,facial expression and poses.Moreover,the experimental results show that the improved method outperform face recognition methods including ECCLDA,CCLDA and LDA.