Relevance Feedback in Image Retrieval Based on RSVM

Support vector machines (SVM) are favored for relevance feedback in content-based image retrieval by utilizing both positive and negative feedbacks. This paper uses incremental reduced support vector machines to get the support vectors and the non-support vectors, then utilizes both positive and negative feedbacks for image retrieval based on SVM. It needn't use the results of retrieval to train SVM again as traditional method. Experimental results show that the model has good effectiveness.