Aggregative query generation

This paper proposes an aggregative query generation which exploits a media document representation called feature term to create a query from multiple media examples, e.g. images. A feature term denotes an interval of one media feature dimension, such as a bin in colour histogram. This approach (1) can easily accumulate features from multiple query examples to generate an efficient query; (2) enables the exploration of text-based retrieval models for multimedia retrieval. Two criteria, minimised chi² and maximised entropy, are proposed to optimise feature term selection. Two ranking functions, KL divergence and tf-idf based BM25 model, are used for relevance estimation. Experiments on the Corel photo collection demonstrate the effectiveness of feature terms.