Comparison of Local Higher-Order Moment Kernel and Conventional Kernels in SVM for Texture Classification

Authors:
Keisuke Kameyama
Affiliations:
Department of Computer Science, Graduate School of Systems and Information Engineering, University of Tsukuba, Tsukuba, Japan 305-8573
Venue:
Neural Information Processing
Year:
2007

Citing 2
Cited 0

The nature of statistical learning theory

The nature of statistical learning theory
Support-Vector Networks

Machine Learning

Quantified Score

Hi-index	0.00

Visualization

Abstract

Use of local higher-order moment kernel (LHOM kernel) in SVMs for texture classification was investigated by comparing it with SVMs using other conventional kernels. In the experiments, it became clear that SVMs with LHOM kernels achieve better trainability and give stable response to the texture classes when compared with those with conventional kernels. Also, the number of support vectors were kept low which indicates better class separability in the nonlinearly-mapped feature space.