Abstract
Based on minimum within-class scatter support vector machines (MCSVM), a new matrix pattern based MCSSVM (MCSVMmatrix) is presented. Accordingly, it is extended by introducing Mercer's kernels in order to solve the problem of nonlinear decision boundaries, which presents a significant matrix pattern based nonlinear support vector machines: Ker-MCSVMmatrix. The above-mentioned approaches not only keep the merits of MCSVM, but, owing to introducing matrix pattern based within-class scatter matrix into support vector machines, theoretically better solve the singular problem of within-class scatter matrix when small sample size problems are dealt with, reduce the time/place complexity when within-class scatter matrix, its invertible matrix and weight vector ω are calculated. Hence, the classification accuracy is improved to certain extent. Experimental results indicate the above advantages of the proposed methods: both MCSVMmatrixand Ker-MCSVMmatrix.
Original language | English |
---|---|
Pages (from-to) | 5602-5610 |
Number of pages | 9 |
Journal | Applied Soft Computing Journal |
Volume | 11 |
Issue number | 8 |
DOIs | |
Publication status | Published - 1 Dec 2011 |
Keywords
- Face recognition
- Matrix pattern
- SVM
- Within-class scatter matrix
ASJC Scopus subject areas
- Software