Natural language search in image collections

Abstract

It is a challenge for today’s library practice to make the entities in digital images retrievable according to the users demands. In our presentation we wish to introduce a searching method which is able to seek certain elements and their environment in the images by applying a search within them in natural language. One key is to this method is extending the Dublin Core metadata system by a new element or by a classifier of the description element which help to enlist the natural language denominations of the elements in the images and the elements’ positions within the image. Following this we are able to find certain elements then calculate and grade the information value of their environment with the help of a modified version of Shannon’s entropy formula. The application of this method enables us to perform further value calculations for certain details within a particular image which make further, more efficient searches possible.