基于Fisher判别的结构化低秩字典学习算法研究

Research on structured low-rank dictionary learning based on Fisher discrimination

为了进一步提高人脸识别方面的性能,提出了基于Fisher判别的结构化低秩字典学习算法。该算法基于训练样本的标签信息将低秩正则化以及结构化稀疏同时引入到所学习的具有识别力的字典上。在字典学习过程中,首先利用样本的重建误差约束样本与字典之间的关系;其次将Fisher准则应用到稀疏编码过程中,使其编码系数具有识别能力;由于训练样本中的噪声信息会影响字典的识别力,所以在低秩矩阵恢复理论的基础上将低秩正则化应用到字典学习过程中;接着,在字典学习过程中加入了结构化稀疏,使其不丢失结构信息以保证对样本进行最优分类;最后,在AR以及ORL人脸数据库上分别进行实验仿真,实验结果表明该方法在人脸识别方面具有可行性。

In order to improve the performance of face recognition,the algorithm of structured low-rank dictionary learning base on Fisher discrimination is put forward.The algorithm adds low-rank regularization and structured sparse to the discerning dictionary learning based on the label information from training samples.In the process of dictionary learning,the improved algorithm firstly adopts the reconstruction of samples to constrain the relationships between samples and the dictionary;Then,the algorithm applies Fisher discrimination criterion to the coding coefficients of dictionary learning to make the coding coefficients possess discrimination;Because the noise in the training samples can influence the discrimination of the dictionary,the algorithm applies low-rank regularization to the dictionary on the basis of the theory of low-rank matrix recovery;Next,in the process of dictionary learning,structured sparse is imposed to avoid losing structure information and guarantee the optimal classification for samples;At last,experiments are performed on the AR face database and ORL face database respectively.The results show that the proposed algorithm possesses the feasibility in the aspect of face recognition.