PexAce: A Method for Image Metadata Acquisition - Semantic Acquisition Games: Harnessing Manpower for Creating Semantics - page 71

5.3 PexAce Personal: Personal Imagery Metadata
During the deployment of the game, we observed several players during game play-
ing. Besides playing (finding card pairs, annotating, running for points), they were
sometimes simply watching on the images, especially those which contained some-
thing unusual. They were ceasing to play for short whiles, enjoying the images. We
have observed this behavior occasionally during the experiments with the Corel 5K
image dataset, but more often this occurred later, when we used the game just for
showcasing it at various events. Then, rather the general domain images, we used
photos with which many of the players were familiar—past year photos taken on
the same occasion. The game acted also as a image presentation tool. And familiar
images made it more attractive.
This suggested the idea to use the game to annotate not just any images, but images
familiar or personal to the player himself. Also, and more importantly: the purpose
of the game would become useful for the player himself. It would help to acquire
metadata useful in maintenance of personal image repositories (e.g. family photos).
This is an unpopular and tedious task, which at the same time, cannot be currently
done by other means than manual work of the image owner himself. The trouble here
is with the highly specific metadata: names of persons, places or events connected
to the images—something that neither crowd or automated method could provide.
We have examined the practices of users in dealing with personal imagery orga-
nization. As a main source, we used the qualitative study of Vainio et al. [ 3 ], which
analyzes the prevailing practices of multimedia resource creators and users (of mainly
images), in regard to the metadata creation and upkeep. According to this study, users
like to interact (create and edit) with the content (of the images), but not with its meta-
data, although their recognize their importance. Even when they admit, they have
trouble with organizing of their repositories, they are not willing to spent time on
metadata creation. The only common practice is the high-level hierarchical organi-
zation of resources into named albums. Apart from this, the retrieval process relies
on chronological sorting and extensive thumbnail browsing.
For us, this represented an opportunity to bring in an alternative solution for
acquiring personal imagery metadata. The idea of this game was that players would
play the semantics acquisition game with their own images and keep the acquired
metadata for their own sake.
One of the important notions of the Vainio's study was, that users would mostly
welcome specific metadata to describe their resources: person names, places and
events where pictures were taken. Therefore, for our own study, we were especially
interested in the game's performance in acquiring these types of metadata.
To achieve our goal— acquire personal imagery metadata through semantics
acquisition game —we modified the original PexAce into a game called PexAce-
Personal . From the player's perspective, the gameplay was kept in its original form.
The only change was the image dataset loaded into the game—prior to the game
start, the player loads his own image set.