A Novel SVM Algorithm Based on Loop-Symmetrical Division for Multi-Class Classification Problem

A novel SVM method is presented, in which loop-symmetrical division is adopted to solve multi-class classification problem. In the proposed method, the classification of multi-class samples are loop-arranged and symmetrical divided, and an error-correcting codes matrix is constructed. With the constructed codes matrix, the class information of testing samples can be found with the decoded function. Experiments on ORL face database verify the efficiency of the our algorithm.