Karido: A GWAP for telling artworks apart

Creating descriptive labels for pictures is an important task with applications in image retrieval, Web accessibility and computer vision. Automatic creation of such labels is difficult, especially for pictures of artworks. “Games With A Purpose” strive to create image descriptions by harnessing the intelligence of human players. Existing implementations are highly successful in terms of player count and number of collected labels, but hardly create comprehensive tag sets containing both general and specific labels. We propose Karido, a new image labeling game designed to collect more diverse tags. This paper describes the design and implementation of the game, along with an evaluation based on data collected during a trial deployment. The game is compared to an existing image labeling game using the same database of images. Results of this evaluation indicate that Karido collects more diverse labels and is fun to play.