摘要
为了进一步提高基于协从表示的人脸识别系统的性能,在概率协从表示(ProCRC)算法和字典学习的基础上提出了一种基于Gist特征和ProCRC的GL-PCRC人脸识别算法。首先提取每副人脸图像的Gist特征,再把人脸图像的Gist特征采用线性判别算法(LDA)方法投影到最优判别子空间,使得到的LDA特征拥有最小的类内离散度以及最大的类间离散度;然后利用LC-KSVD方法对LDA特征进行迭代训练从而得到新的学习字典;继而通过ProCRC算法快速得到稀疏系数;最后通过计算测试样本属于各个类别的概率进行分类。分别在ORL和扩展的YaleB人脸库上进行实验检测的结果表明,与传统的协从表示方法相比,本文给出的方案可以使人脸识别系统的性能得到显著的提升。
In order to improve the performance of face recognition system based on collaborative representation,this paper proposes a face recognitional algorithm based on Gist feature and probabil istic collaborative representation (ProCRC).Firstly, it extracts the Gist feature of each face image,and projects them to optimal d iscriminant subspace by using the linear discriminant analysis LDA method,which can ensure that the LDA feature has the smallest cla ss scatter and maximum between class scatter. Then,it obtains the new learning dictionary by iteratively training the LDA fea ture using the LC-KSVD method,and the sparse coefficient is obtained by the ProCRC method.At last,it classifies them by calculating the probability that t he test sample belongs to each category. Experimental results on the ORL and extended YaleB database show that the face r ecognition rate can be significantly improved compared with the traditio nal collaborative representation.
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2017年第12期1365-1371,共7页
Journal of Optoelectronics·Laser
基金
山东科技大学教学研究(JG201506)
山东科技大学研究生教育创新(KDYC13026
KDYC15019)
山东省研究生教育创新计划(01040105305)资助项目
关键词
人脸识别
协从表示
Gist特征
字典学习
face recognition
collaborative representation
Gist feature
dictionary learning