CAJ | 학술논문

针对传统谱算法在人脸识别中的局限，该文提出一种基于改进高斯过程隐变量模型(GP-LVM)的多角度人脸识别算法。首先，通过高斯过程(GP)对人脸流形建立概率模型，得到高斯过程隐变量模型(GP-LVM)；其次，分析GP-LVM得到共有信息(shared information)和独有信息(private information)，利用概率最大化与拉格朗日乘子法得到参照矩阵和参照值；最后，实现多角度人脸识别。选取Yale, JAFFE, FERET, CMU-PIE 4类数据集进行对比实验，实验结果表明：该文提出的算法可以有效地识别多角度人脸，针对无角度人脸识别也具有良好的效果。
침대전통보산법재인검식별중적국한，해문제출일충기우개진고사과정은변량모형(GP-LVM)적다각도인검식별산법。수선，통과고사과정(GP)대인검류형건립개솔모형，득도고사과정은변량모형(GP-LVM)；기차，분석GP-LVM득도공유신식(shared information)화독유신식(private information)，이용개솔최대화여랍격랑일승자법득도삼조구진화삼조치；최후，실현다각도인검식별。선취Yale, JAFFE, FERET, CMU-PIE 4류수거집진행대비실험，실험결과표명：해문제출적산법가이유효지식별다각도인검，침대무각도인검식별야구유량호적효과。
The traditional spectrum algorithms are limited in face recognition issue. For its characteristics of issue, a novel multi-angle face recognition method based on modified Gaussian Process Latent Variable Mode (GP-LVM) is proposed. Firstly, the probabilistic model of face manifold is established with the Gaussian Process (GP), and the GP-LVM can be gotten. Secondly, the shared information and private information can be gotten by analyzing the GP-LVM. Thereafter, the reference matrices and the reference values are calculated with maximum probability and Lagrange algorithm. Finally, the multi-angle face recognition can be achieved. The four classes of data sets are selected as the experimental data, which consist of Yale, JAFFE, FERET and CMU-PIE. The experiment results show that the proposed method not only has a great effect to recognize multi-angle face, but it can be applied to no angle face recognition.