基于L1范数的核判别保局投影算法研究-计算机技术专业论文.docxVIP

安徽大学硕士论文 安徽大学硕士论文 ABSTRACT ABSTRACT In recent years，with the development of new techniques and tools which for collecting information，the information we get in real world always in high dimension．So dimensionality reduction technique is becoming the key technology in processing high—dimensional data and there are more and more dimensionality reduction methods．Principal component analysis(PCA)and linear discriminant analysis(LDA)are two popular classical linear dimensionality reduction methods． Discriminant locality preserving projections based on L 1-norm maximization (DLPP-L1)is also an efficient linear dimensionality reduction technique．DLPP-L 1 preserving the local neighborhood structure and using the discriminant information of original samples．DLPP-L1 attempts to capture an optimal transformation matrix which makes the samples in same class as compact as possible and meanwhile the samples in different class as separate as possible．With the successful of the kemel function used in SVM．the methods based on kernel function has attracted more and more attention．It mapped the data from original space to high—dimensional space and make the data more distinction with each other in high—dimensional space Researchers introduced kernel function into linear dimensionality reduction methods． The experimental results shows that the dimensionality reduction method based on kernel function has obviously increased in performance．Inspired by this，in this paper，we proposed a new method named as kemel discriminant locality preserving projections based on L 1一norm maximization(KDLPP—L 1)．Meanwhile the methods use input images matrix directly instead of transformed them into high—dimensional vector which is avoiding the problem of dimension disaster have achieved great performance．Inspired by this，the second work of this paper is to proposed new method that use input images matrix directly termed as two—dimensional discriminant locality preserving proj ections based on L 1-norm m