Evaluating image browsers using structured annotation

In this article we address the problem of benchmarking image browsers. Image browsers are systems that help the user in finding an image from scratch, as opposed to query by example (QBE), where an example image is needed. The existence of different search paradigms for image browsers makes it difficult to compare image browsers. Currently, the only admissible way of evaluation is by conducting large‐scale user studies. This makes it difficult to use such an evaluation as a tool for improving browsing systems. As a solution, we propose an automatic image browser benchmark that uses structured text annotation of the image collection for the simulation of the user's needs. We apply such a benchmark on an example system.