Boost SVM active learning for content-based image retrieval

Content-based image retrieval (CBIR) can be viewed as a classification problem, and the classical support vector machine active learning (SVM_Active) algorithm gives a satisfactory solution. In this paper, based on the SVM_Active algorithm, our contribution is: boosting method is incorporated with SVM_Active to get the Boost SVM_Active (BSVM_Active) algorithm. Following the basic sample re-weighting idea of AdaBoost, we modify this method to be adaptive to CBIR problem. Boosting method can improve the performance of SVM_Active classifier with both higher accuracy and faster training process. Experiment results over three different scales datasets show that our new method can achieve consistently higher performance than original SVM_Active.