Bayesian approaches to content-based image retrieval

Abstract

This thesis addresses some issues in the relatively new field of Content-Based Image Retrieval. Content-based image retrieval is a technique that uses the visual content of images to aid searches from large scale image databases. The field of content-based image retrieval is growing in importance as image archives grow in size in many fields, and a means to search for visual content on the Web. In this thesis we investigate Bayesian approaches to content-based image retrieval. Starting from the work of Cox et al. (1996) and Cox et al. (2000), we propose a retrieval system that attempts to capture properties of the visual content of images and how a user conducts a search. Decision theory is applied to the problem of selecting images to retrieve. The system is evaluated using a variety of tests with users.