Some results in image representation

Here are some results of the re-synthesis of continuous image representations I am working on. Results are roughly in chronological order (you hopefully should see an increase in the quality/difficulty of the results ;-).

Some early examples

A first image:

and its reconstruction:

The result is also available as a PaVRML definition (download here).

First, the representation is vectorial. This means that the reconstructed image can be at any resolution (hence the larger image). Second, the representation is region-based (which is clearly visible in the PaVRML file, see above). This means that we can remove some parts of the original image, e.g. the background:

Some more fancy effects can also be done, using an image rendering and compositing software developed at the University of Bath, like changing the lighting (of the background here):

More recent examples

The following image pairs show respectively the original image and its re-synthesised version.

Even more recent examples

Here, we show an image of a truck, an image made from the truck only, and an image containing the goose and the truck.