融合字典学习的小样本人脸鉴别

基于稀疏表示的人脸鉴别方法通过提高字典的判别力来提高识别准确率,本文针对小样本训练,提出一种新的融合字典学习方法.首先利用Fisher判别准则及LBP金字塔进行数据预处理;其次提出新的融合字典学习模型,该模型由公共字典、类别特色字典及扰动字典三部分构成,分别提取数据共性、不同类别数据的特殊性以及异常情况下的数据扰动性;最后根据融合字典模型提出一种新的分类器,并在AR、YALE、CMU-PIE、LFW人脸数据库进行实验,结果表明本文算法具有更高的识别率和有效性.

Small Sample Face Recognition Based on Fusion Dictionary Learning

Face recognition based on sparse representation improves recognition accuracy by improving the discriminative power of dictionaries.This paper proposes a fusion dictionary learning algorithm.Firstly,Fisher criterion and LBP pyramid are used to preprocess data.Then,based on the features of the data,a fusion dictionary learning model is proposed to comprehensively study the common dictionary,class-specific dictionary,and perturbation dictionary to extract the commonness between the data,the particularity of the different types of data,and the data perturbation under abnormal conditions.Finally,a new classifier is proposed based on the model,and experiments are performed on face databases such as AR,YALE,CMU-PIE,LFW.The results show that this algorithm has higher recognition rate and effectiveness.