A NEW SVM KERNEL FOR TEXTURE CLASSIFICATION (TA-P4)

Author(s) : Mahdi Sabri (Ryerson University, Canada) Javad Alirezaie (Ryerson University, Canada)

Abstract : The performance of the Support Vector Machine(SVM) algorithm is highly dependent on the choice of the kernel function suited to the problem at hand. In a Support Vector Machine algorithm feature selection is implicitly performed by kernel function. On the other hand, feature selection is the most important stage in any texture classification algorithm. In this work, the performance of SVM is improved by choosing an optimized space-frequency (SFR) kernel function. The proposed method is evaluated in two texture problem as well as multi-texture problem. The results are compared with the original SVM and some other recently published texture classification methods. The comparison shows a significant improvement in error rates( more than 40\% in compare with original SVM and more than 60\% in compare with two other methods).

Menu