Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Research on New Methods of Linear Discriminant Analysis and Their Applications

Author LiDaoHong
Tutor ChenSongCan
School Nanjing University of Aeronautics and Astronautics
Course Applied Computer Technology
Keywords Pattern Recognition Linear discriminant analysis Rank restrictions Matrix singularity Threshold selection Binarization Image processing
CLC TP391.41
Type Master's thesis
Year 2005
Downloads 550
Quotes 5
Download Dissertation

Fisher linear discriminant analysis (FLDA, or LDA) is one of the very effective feature extraction method has been widely and successfully applied in the field of dimensionality reduction , data analysis and pattern classification . Conducted a series of studies on linear discriminant analysis and its application . First of all , on the basis of the multi-feature linear discriminant analysis (Multi-feature LDA, MFLDA) , nuclear - based multi-feature linear discriminant analysis ( kernel MFLDA kMFLDA ) to overcome the multi-feature linear discriminant analysis can not be effectively solve nonlinear separable problem is inadequate; Secondly, the rank restrictions exist in the traditional linear discriminant analysis and singularity problem by amending the original LDA criteria , proposed a modified linear discriminant analysis criteria (Modified LDA , ModLDA), in order to overcome the the rank limitations and relaxation of the restrictions of the singularity problem . The experimental results show that this paper, two linear discriminant analysis method is superior to the traditional LDA or equivalent effect in a number of international standard data sets and artificial datasets . Finally, linear discriminant analysis conducted applied research in the gray image binarization problem . Grayscale image objects usually have a uniform grayscale value , while the background generally has a non-uniform gray value features , from human intuition of the observed image , the design of a new two types of criterion , by optimizing the the criteria to obtain the optimal threshold value , and then use this threshold value to achieve a gray image binarization . The comparison experiments show that with the other methods : image segmentation obtained by this method has a better visual effect .

Related Dissertations
More Dissertations