Abstract: GEPSVM is a newly proposed binary SVM in recent years. It learns two optimal hyperplanes by solving the generalized eigenequation and determines the categories of patterns based on the distances between test sample and the two hyperplanes. Compared with traditional SVM, GEPSVM can reduce the time complexity, but it still leaves singulari- ty issues unsolved. This paper introduced a new algorithm丁DMSVM(Twin Distance of Minimum and Maximum Sup- port Vector Machinc)which seeks two optimal hyperplanes through solving the standard eigenequation and requires the minimal average distance between the same class of samples and hyperplanes. Compared with GEPSVM, I}DMSVM has the following advantages:it further reduces the time complexity and does not rectuire the introduction of regular items which improves the generalization performance and avoids singularity.

Key words: Pattern recognition,Eigenvector,Support vector machine,I_agrange multiplier method