An abstract content-based image retrieval system based on activity theory

Traditional methods of image retrieval require that meta-data is associated with the image, commonly known as keywords. These methods power many World Wide Web search engines and accomplish reasonable amounts of search accuracy. Though some content based image retrieval (CBIR) systems use both semantic and primitive attributes to match search criteria, history has proven that it is difficult to extract linguistic information from a 2D image. In this research, activity theory is used as a base to demonstrate how semantic information can be retrieved from objects identified in an image. Using an image segmentation technique by The Berkeley Digital Library Project (Blobworld), and combining it with object-to-community relationships, a high-level understanding of the image can be demonstrated.