An Improved HIK for Object Categorization

Abstract

In this paper we applied a Histogram Intersection Kernel (HIK) method for categorization of the Caltech101 dataset. We analyzed the principles of HIK and propose an optimal linear combination of kernels used in Spatial Pyramid model (SPM). Sift algorithm is utilized to detect and describe image features based on Bag of Words model. The performance is compared between HIK and general RBF using SVM for the classification. The experimental results show that, based on the same image dataset, HIK outperforms RBF. Furthermore, HIK-SVM's performance is improved with the increasing layers of SPM. On the contrary, RBF-SVM's performance worsens when the layers of SPM increase.