基于线性鉴别的无参数局部保持投影算法

Parameter-free locality preserving projection based on linear discriminant

针对局部保持投影算法的无监督性质和参数选择复杂性问题,结合线性鉴别分析算法,提出一种改进的有监督无参数局部保持投影算法(Linear Discriminant Supervised Parameter-free Locality Preserving Projection algorithm,LD-SPLPP). LD-SPLPP算法采用监督模式并使用广义Dice系数的方法构建近邻矩阵,有效避免LPP(Locality Preserving Projection)算法参数选择调整的问题.新算法在UCI的八个低维度数据集和两个高维度人脸数据库上进行了实验,通过对数据的特征提取,采用最近邻分类法统计识别率,并分析了实验分类后的数据值与算法性能的关系.上述实验过程中,将新算法与PCA,LDA,ULDA,OLDA,LPP,SPLPP,PSKLPP,PSLMM和EP-SLPP算法进行了对比,实验结果证明了LD-SPLPP在数据降维和特征提取方面的有效性.

In this paper,considering the character of unsupervised and complexity of parameter selection of the locality preserving projection algorithm,an improved Linear Discriminant Supervised Parameter-free Locality Preserving Projection(LD-SPLPP)algorithm was proposed. In order to avoid the problems of parameters selection and adjustment of Locality Preserving Projection(LPP)algorithm,LD-SPLPP constructs an affinity matrix under the supervised mode and uses generalized Dice coefficient. LD-SPLPP algorithm performed experiments based on eight kinds of low-dimension datasets of UCI and two kinds of high-dimension human face databases. LD-SPLPP algorithm carried out the feature extraction on the data,used nearest neighbor classifier to get correct recognition rate and analyzed the relationship between the value of the classified data and the performance of the algorithm. During the experiments,LD-SPLPP is compared with PCA,LDA,ULDA,OLDA,LPP,SPLPP,PSKLPP,PSLMM and EP-SLPP,and the experimental results demonstrate that the proposed method is effective on the feature extraction.