Support Vector Machine Concept-Dependent Active Learning for Image Retrieval

Relevance feedback is a critical component when designing image databases. With these databases it is difficult to specify queries directly and explicitly. Relevance feedback interactively learns a user’s desired output or query concept by asking the user whether certain proposed images are relevant or not. For a learning algorithm to be effective, it must learn a user’s query concept accurately and quickly, while also asking the user to label only a small number of images. In addition, the concept-learning algorithm should consider the complexity a concept in determining its learning strategies. In this paper, we present the use of support vector machines active learning in a concept-dependent way (SVM Active) for conducting relevance feedback. We characterize a concept’s complexity using three measures: hit-rate, isolation and diversity. To reduce concept complexity so as to improve concept learnability, we propose a multimodal learning approach that uses images’ semantic labels to intelligently adjust the sampling strategy and the sampling pool of SVM Active . Our empirical study on several datasets shows that active learning outperforms traditional passive learning, and concept-dependent learning is superior to the traditional conceptindependent relevance-feedback schemes.