« Web 2.0 meets the National Library of Scotland's digital library | Main | OAIS compliancy »
August 26, 2008
Some comments on image tagging and organization
An upcoming exhibit at the Museum of Contemporary Art in Antwerp entitled "The Order of Things" confronts some of the questions on image archives and online image banks. Natural language searching is often faulted its non-specificity and inability to produce fully relevant search results. An interesting juxtaposition when combined with the visual image. Order and classification often lose out to rather arbitrary category assignments by the creator. These can make for an interesting cultural and social commentary on the varying associations and significance of images, though also creates many problems in the search.
In a book written in 1966 by Michael Foucault with the same title as the exhibition, the author writes about the will to order by the individual. "Order, on the other hand, is established without reference to an exterior unit..... one cannot know the order of things in their isolated nature, but by discovering that which is the simplest, then that which is next simplest, one can progress inevitably to the most complex things of all" (p. 59). Foucault identifies a problem that is inherent to the classification of an individual image. The image is considered and referred to in a relatively simple and flat dimension, as the thing itself, often not as a series or in any relation to other items. These methods of classification also usually occur at different times and within different environments, so is it a surprise that there are so many error and mistakes in searching with natural language, particularly with images?
In indexing images, there is also a difference of thinking about the content over the concept (or vice versa). If you were to add key terms to a work of art, for example, a difference between the 'ofness' and 'aboutness' of the work will quickly become apparent. Is the content of the work important here, or other iconological identifiers (cultural, religious, symbolism)? Sometimes it depends on the purpose of the database, or even in the user group who will be using the database. Some art databases are structure to search using color, shape, texture, material or other aspects of the work that would not be useful outside of that context.
A related work by Roy Arden, 'The World as Will and Representation' (2007) is a collection of 28,144 images combined in a whirlwind slide show format of images found online and other sources. The collection of images illustrates the chaotic nature in the multitude of images that were taken from online images. These images articulate the disorder and lack of context when separated from original environment.
Google has asked for help on their Image Labeler, where people provide labels to different images which are compared to an identical set of images from another person. People are asked to focus on descriptive labeling in this exercise. These tend to become generic, and less specific to place, time and content. Instead the basic elements are chosen ('guitar', 'singer', 'bird', 'sky', etc.)
ALIPR is an automatic photo tagging and visual image search. The automatic image annotation is limited to only 322 English words at the moment, though also points to more generic tagging. Interestingly, the engine includes emotions in this limited set (some examples are 'surprising', 'amusing', 'pleasing', etc.). As more images are added, the results become more broad and less specific, and therein more time spent sifting through unrelated hits.
Posted by vad17 at August 26, 2008 03:35 PM
Trackback Pings
TrackBack URL for this entry:
http://blog.case.edu/digitallibrary/mt-tb.cgi/18567