Image Semantic Classification Based on Integration of Feature Subspaces

This paper proposes an image semantic classification algorithm based on feature subspaces. It is implemented by SVM and AdaBoost algorithm. In every feature subspace, a SVM is trained. According to the error rate of every SVM, the integrating weight of feature subspace is determined, with which different subspace features are concatenated into a feature vector. Then AdaBoost algorithm is employed to train classifier, whose weak classifiers are SVMs. Test and comparison are conducted between the proposed algorithm and the linear kernel SVM classifier, the RBF-SVM classifier as well as the AdaBoost classifier. The proposed algorithm is turned out to be the best one. In addition, toward the feature subspace algorithm proposed, this paper has also given the comparative experiment, its contrast group is respectively color feature space, textural feature space and equal-weight color and texture space. The experiment shows that this method increases the classifier precision effectively.