The datasets on (and under) this page were put together by Renato O. Stehling (in collaboration with Mario Nascimento, and Alexandre X. Falcao), and have been used for evaluating content-based image retrieval (CBIR) techniques.

The images below are the query images we have used. To view the associated (expected) answer sets click on the images. The 15 answer sets have a total of 178 images, thus an average of about 11 images per set. The images have been resized to facilitate their viewing, i.e., they are not the original image sizes as found in Corel's CDROM.