Memorization and Semi-Supervision Based Active Relevance Feedback Algorithm for Content-Based Image Retrieval

To improve the efficiency of relevance feedback quickly,an integrated memorization and(semi-)supervision active relevance feedback algorithm is presented.In its early stage,more positive samples are obtained through memorization.The problem of biased training samples is solved efficiently through labeled and unlabeled training samples and accurate initial SVM classifier is obtained;In the later stage,samples required for labeling by users reduced largely and convergent rate improved greatly by the active learning algorithm which selects the most useful samples in database to solicit the user for labeling.Experimental results on 5?000 Corel images library showed that the proposed algorithm can greatly improve the efficiency and accuracy and converge to user's query concept quickly.