共享空间基-逐类剩余空间基混合稀疏表示人脸识别算法

摘要

In view of the ordinary sample training dictionary learning ' s deficiency of unusing class common information , an algorithm for sparse representation of face recognition using shared space basis -class wise remaining space basis hybridization is proposed based on introducing shared space and remaining space associated with classes .The algo-rithm extracts the PCA ( principal component analysis ) features of training samples , and acquires the unlabeled shared space basis and its reconstruction samples to obtain class common information ; then , obtains a differential training set by combining the original samples , and constructs a class specific remaining basis by introduing inter class difference information;and finally , obtains the hybrid dictionary by combining the shared space common basis and the basis of class-wise remaining space , and then classfies the testing samples by using of residual error SRC ( sparse representation classification ) criterion.This method not only makes full use of the orthogonal property of the hybrid dictionary , but also gives full play to the discriminating ability of the remaining space and the role of sparse approximation of shared information , making the close combination of structured dictionaries with pattern classification.The results of the experiments on facial databases of AR ,CMU PIE,Extended Yale B verify the pro-posed algorithm ' s effectiveness .%针对传统训练样本字典学习未利用类共有信息的不足,引入共享空间和与类别相关的剩余空间,提出了共享空间基-逐类剩余空间基混合稀疏表示人脸识别的算法.该算法首先提取训练样本主成分分析(PCA)特征,获取无标记的共享空间基及其重构样本得到类共有信息;然后结合原始样本得到差分训练集合,并引入类间差异信息构建逐类特异性剩余空间基;最后融合共享空间基和剩余空间基,利用残差判别函数完成模式分类.该方法不仅利用混合空间的正交特性,而且发挥剩余空间的鉴别能力和共享信息稀疏逼近的作用,使结构性字典和模式分类紧密结合.该方法的有效性,分别通过用AR、CMU PIE、Extended Yale B人脸数据库进行的实验得到验证.