CAJ | 학술논문

邻域保持嵌入是局部线性嵌入的线性近似,强调保持数据流形的局部结构.改进的最大间隔准则重视数据流形的判别和几何结构,提高了对数据的分类性能.文中提出的核岭回归的邻域保持最大间隔分析既保持流形的局部结构,又使不同类别的数据保持最大间隔,以此构建算法的目标函数.为了解决数据流形高度非线性化的问题,算法采用核岭回归计算特征空间的变换矩阵.先求解数据样本在核子空间中降维映射的结果,再解得核子空间.在标准人脸数据库上的实验表明该算法正确有效,并且识别性能优于普通的流形学习算法.
린역보지감입시국부선성감입적선성근사,강조보지수거류형적국부결구.개진적최대간격준칙중시수거류형적판별화궤하결구,제고료대수거적분류성능.문중제출적핵령회귀적린역보지최대간격분석기보지류형적국부결구,우사불동유별적수거보지최대간격,이차구건산법적목표함수.위료해결수거류형고도비선성화적문제,산법채용핵령회귀계산특정공간적변환구진.선구해수거양본재핵자공간중강유영사적결과,재해득핵자공간.재표준인검수거고상적실험표명해산법정학유효,병차식별성능우우보통적류형학습산법.
Neighborhood preserving embedding is a linear approximation to locally linear embedding,and it emphasizes preserving the local structure of the data manifold.The modified maximal margin criterion focuses on the discriminant and geometrical structure of the data manifold,and it improves the classification performance of the data.An algorithm is proposed called neighborhood preserving maximal margin analysis of kernel ridge regression.It preserves the local structure of the manifold and maximizes margins between the data of different classes to construct the objective function.As the data manifold is highly nonlinear,the kernel ridge regression is adopted to calculate the transformation matrix.The mapped results of the data samples are obtained by the proposed algorithm in the kernel subspace firstly,then the kernel subspace is obtained.The experimental results on the standard face database demonstrate that the proposed algorithm is correct and effective.Moreover,it achieves better performance than the popular manifold learning algorithms.