CAJ | 학술논문

针对完备鉴别局部保持投影算法所求得的最优判别矢量间存在信息冗余问题，提出了核的正交完备鉴别局部保持投影算法。通过将核函数技术与正交性原理融合，采用高斯核函数将原始样本映射到高维特征空间，在高维特征空间的局部总体散度矩阵中计算最优判别矢量，只需在整个范围内对值域空间进行特征值分解，去除局部零空间达到样本降维目的。该算法分别在 UMIST 人脸库和 JAFFE 人脸表情库上进行实验，实验结果表明算法的识别率高达95．59％。
침대완비감별국부보지투영산법소구득적최우판별시량간존재신식용여문제，제출료핵적정교완비감별국부보지투영산법。통과장핵함수기술여정교성원리융합，채용고사핵함수장원시양본영사도고유특정공간，재고유특정공간적국부총체산도구진중계산최우판별시량，지수재정개범위내대치역공간진행특정치분해，거제국부령공간체도양본강유목적。해산법분별재 UMIST 인검고화 JAFFE 인검표정고상진행실험，실험결과표명산법적식별솔고체95．59％。
This paper proposed the method of the orthogonal complete discriminant locality preserving projections of kernel, which was aimed at the redundant information of the optimal discriminant vectors by the method of complete discriminant locali-ty preserving projections.Combined the convergence kernel technology with orthogonal principle,it mapped the original sample by the method of Gauss kernel function to the high-dimension feature space,and calculated the optimal discriminant vectors in the locality total scatter matrix of high-dimension feature space by the eigenvalue decomposition in the whole range space.It was feasible to remove the locality zero space for the dimensionality reduction.The algorithm made the experiment on UMIST face database and JAFFE facial expression database.Experimental results show that the recognition rate can reach more than 95.59%.