Beyond 2D-grids : a dependence maximization view on image browsing
Novi Quadrianto, Kristian Kersting, Tinne Tuytelaars and Wray Buntine
In: 11th ACM SIGMM international conference on multimedia information retrieval - MIR 2010, 29-31 March, 2010, Philadelphia, USA.
Ideally, one would like to perform image search using an intuitive
and friendly approach. Many existing search engines
return a set of pages, where each page contains several images
ranked based on how relevant they are to the query.
While this certainly has its advantages, arguably, a more intuitive
way to present the image search result is to organize
images based on their similarity, for example, according to
color or semantics, into arbitrary structures such as hierarchies
and spheres. This paper focuses on designing such a
navigation system for image browsers. The design uses a
recently developed machine learning technique, called kernelized
sorting. Kernelized sorting is a general technique
for matching pairs of objects from dierent domains without
requiring the cross-domain similarity measure and hence
allows sorting images into arbitrary structures. Moreover,
some images can be preselected for instance forming the tip
of the hierarchy allowing to subsequently navigate through
the search results in the lower levels in an intuitive way.