New research published in IEEE Transactions on Pattern Analysis and Machine Intelligence uses dolphin photos from the SCUBA Travel site to show, for the first time, how to build 3D morphable models from 2D photographs.
You can watch the video of the system in action at http://www.people.usi.ch/cashmant/dolphins.htm.
Morphable models of the human head and body have enabled a myriad of applications in computer graphics, special effects, and computational photography. However, building a morphable model for a new object category is currently a task that demands considerable effort and often requires full 3D scans
For live animals, the difficulty is even greater: obtaining the 3D scans is essentially impossible even using today’s real-time depth sensors, particularly if the animal is moving – for example a dolphin in midleap.
In one sense, however, there is no shortage of data: you can obtain dozens of high-quality images of the target animal simply using a search engine. The question is, whether such a collection contains enough information to build a 3D morphable model of the animal.
The authors of the paper, Thomas J. Cashman and Andrew W. Fitzgibbon, show that given 32 images of a dolphin they can generate a morphable model that looks like a 3D dolphin.
What shape are dolphins? Building 3D morphable models from 2D images, IEEE Transactions on Pattern Analysis and Machine Intelligence .