Content-Aware Projection for Tiny Planets

Abstract

Tiny Planets visualise the world looking down at the ground, with physically unrealisable projections that curve the ground plane to look like small worlds. Whilst certain geometries, such as Stereographic, are known to give good Tiny Planet visualisations, the best projection to use depends on the image content. In this work we define a family of Tiny Planet projections that includes several commonly used projection types, but allows for datadependent adaptation to best present the image content to the viewer.We show how to select optimal content-aware projections from this set, minimising distortions from conformality whilst closing gaps and emphasising salient areas in the scene.