Image Annotation: Upgrading a Picture from Pixels to Knowledge

1. Challenge

As the number of images in digital archives grows, an ability to comprehensively describe content of an image becomes crucial. Search engines should recognize in some way what is shown on the picture in order to find the desired image. At the same time, human users should be also provided with complete information about image content so they could retrieve maximum knowledge from the image.

In the ideal situation, each important element visualized on the image would be accompanied with an individual description stored in a way that would enable: (a) clear identification of the object to which it refers; (b) indexing by search engines in order to gather semantic information about individual elements or parts of the image; (c) referring to these individual objects rather than to the whole image when the image is found; (d) providing complete information about every important element of the image to users.

Describing individual elements or areas of image content is known as image annotation. There are various approaches that try to make work with image content more effective both for personal and professional use. This article discusses their strengths and weaknesses to help information developers, documentation managers, librarians, and other people involved in content management choose the right approach that can fit their needs.

While evaluating the approaches to image annotation, the following criteria should be taken into consideration:

Clear identification. When it comes to identification of individual objects on an image, one of the most difficult tasks is development of a referencing system that will enable a viewer to clearly understand to which element the textual comment corresponds. If a picture contains multiple objects (for example, group photos) or compound objects (for example, architectural ensemble), clear identification is the matter of understanding the photo.

Availability of the original вЂњpureвЂќ view. It is hardly to believe that someone would agree to spoil original pictures by putting text directly on an image without an ability to hide it. It is especially crucial for images that have great value (for example, old family photos) and should remain intact.

Transferability. Although the meta-data formats described above enable describing the image only on the top level, their main advantage is an ability to share the image with meta-information about its content. It is not lost wherever the image goes. Object-specific textual annotation should be also stored in a way that ensures distributing annotation with an image regardless specific application used for viewing the picture. For example, if annotation is added by using a special program or online service, it should be available everywhere rather than on this specific program or Web site.

Searchability. To get more relevant search results, (a) an ability to index data about individual elements of the image should be provided; and (b) search should be available not only within specific application or Web site.