An Improved Method For Support Vector Machine-based Active Feedback

Relevance feedback schmes based on support vector machines (SVM) have been widely used in content-based image retrieval (CBIR). SVM-based relevance feedback has often bad performance when the number of labeled positive feedback samples is small. This paper presents a method to use the unlabeled data to improve the performance of SVM classifier, which has only a few labeled training examples. We adapt an improved active learning approach to select most informative data from the unlabeled samples set. It can reduce to compute some unnecessary information for feedback results and only label few samples. It can be used in pervative computing availably. Experiments show our approach can use the unlabeled samples effectively, reduce to label more unnecessary data, and improve the classifier´s performance.