Konferensartikel

In this paper, we introduce a projection technique that aims to place points representing individual images in a two-dimensional visualization space so that proximity in this space reflects some sort of similarity between the images. This visualization technique enables users to employ their visual ability to evaluate the significance of metadata as well as the characteristics of classification methods and distance functions. It can also be used to recognize and analyze patterns in large sets of images, and to get an overview of the entire body of pictures from a given set. The projection technique only uses a similarity function for calculating a suitable distribution of the points in the visualization space and has a linear time complexity.

2D visualization; information search and retrieval; clustering