Computational Photography Project 5 - afarley

Into the Picture


The idea of changing the veiw point of a picture is a powerful one, but of course assumptions must be made. The assumptions of the TIP paper by Horry et al. is that the majority of the image can be modeled by 5 polygons arranged in a box like configuration. In short it is very much like the camera mapping technique widely used in special effects with the addition of automated alpha matte elements.

Ancient Village

A nice corridor. Aside from the far temple on the left it worked quite nicely. It does break up once close to the far wall since buildings to the side end up mapping to the the top plane.
original:

new:


And the brilliant derived textures:

Gondola Venice

This was slightly harder to get a novel image since the boat was too close to include in the far plane and thus is badly smeared.
old:

new:

Salisbury

The arches clearly are not able to be completly expressed with a simple box projection, but the results are still fairly good. Note that the curvature was highly veiw dependent therefor animation would be out of the question.
old:

new:

St. Jerome

This picture was near ideal and had predictable results in even more extreme positions. The lion of course was slightly stretched.
old:

new:

And something a bit more extreme: