Multi-View Sampling for Relevance Feedback in Image Retrieval

Labelling is a boring task for users in relevance feedback. How to maximally reduce the labelling is crucial for relevance feedback algorithms. In spirited by active learning and co-testing, we proposed a Co-SVM algorithm to improve the efficiency and effectiveness of selective sampling in image retrieval. In Co-SVM, color and texture are looked as sufficient and uncorrelated views of an image. SVM classifier is learned in color and texture feature subspaces, respectively. Then the two classifiers are used to classify the unlabelled data. These unlabelled samples that disagree in the two classifiers are chose to label. The experimental results show that the proposed algorithm is beneficial to image retrieval