Sifting through Images with Multinomial Relevance Feedback

This paper presents the theory, design principles, implementation and performance results of a content-based image retrieval system based on multinomial relevance feedback. The system relies on an interactive search paradigm in which at each round a user is presented with a set of k images and is required to select one that is closest to their target. Performance is measured by the number of rounds needed to identify a specific target image as well as the the average distance from the target of the set of k images presented to the user at each iteration. Results of experiments involving simulations as well as real users are presented. The conjugate prior Dirichlet distribution is used to model the problem motivating an algorithm that trades exploration and exploitation in presenting the images in each round. A sparse data representation makes the algorithm scalable. Experimental results show that the new approach compares favourably with previous work.