The development of image retrieval methods and tools that efficiently and accurately search for images based on their content is an interesting challenge.

This paper describes a platform with a multiagent architecture that is able to retrieve images similar to a given query image. A set of agents is developed, each of which is able to compare the query image with a given database of images according to a different criterion, and then select the most similar ones. This comparison is dependent on the results of a previous importation process in which each agent extracts the necessary features from the images imported to the database. Although this preprocessing of images is very time consuming, it is performed prior to retrieval, therefore allowing the retrieval process to be efficient.

The final set of selected images is obtained by a voting system that integrates the contributions coming from different agents; three voting methodologies are implemented.

Finally, results demonstrating the performance of the system in terms of precision and recall are shown. Even though the system seems to work well enough, some questions arise about the proposed solution. The first question is related to the need for preprocessing of the image base, which prevents the solution from being applicable when images are not available in advance. Secondly, presenting the system’s architecture as a multiagent one is arguable, since the so-called agents do not seem to exhibit any of the distinctive properties of this kind of software entity; a set of distributed objects would probably do.

The paper is well structured and written in a clear language. Despite the mathematical apparatus, nonexpert readers will be able to follow the argument.