期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2014
卷号：63
期号：3
出版社：Journal of Theoretical and Applied
摘要：Scene classification, the classification of images into semantic categories is a challenging and important problem nowadays. We present a procedure to classify real world scenes in eight semantic groups of coast, forest, mountain, open country, street, tall building, highway and inside city using support vector machines. Traditional classification approaches generalize poorly on image classification tasks, when the classes are non-separable. In this paper we used Support Vector Machine (SVM) for scene classification. SVM is a supervised classification technique, has its roots in Statistical Learning Theory and have gained prominence because they are robust, accurate and are effective even when using a small training sample. By their nature SVMs are essentially binary classifiers, however, they can be adapted to handle the multiple classification tasks common in scene classification. This paper shows that support vector machines (SVM�s) can generalize well on difficult scene classification problems. SVMs can efficiently perform a non-linear classification using kernel trick, implicitly mapping their inputs into high-dimensional feature spaces. In this paper 4 types of kernels (linear, polynomial, gaussian and sigmoidal kernels) are used with support vector machines. It is observed that Gaussian kernel outperform other types of kernels. Moreover, we observed that a simple remapping of the input x to x' improves the performance of linear SVM�s to such an extent that it makes them, for this problem, a valid alternative to RBF kernels.
关键词：Support vector machine; Kernel; cross validation; dimensionality reduction; linear kernel; polynomial kernel; gaussian kernel and sigmoidal kernel
