3D volume morphing

This page shows some results from the volume morphing branch of the volume rendering project of the Stanford computer graphics laboratory.

Notes:

The original figures (in losslessly compressed TIFF format) can be retrieved by clicking on the corresponding thumbnail.
All the renderings shown here, including both static images and movies, are volume renderings, not polygon renderings.
The following MPEG movies do not capture the full quality of the morphs.

Human to orangutan

Description	Source	Halfway morph	Target
The human head on the left is transformed into the orangutan head on the right; the halfway morph is shown in the middle. Both volumes were obtained by CT scans.
Movies	Human head (892KB).	Morph sequence (570KB).	Orangutan head (836KB).

Dart to X-29

Description	Source	Halfway morph	Target
The dart on the left is transformed into the X-29 on the right; the halfway morph is shown in the middle. Both original volumes were obtained by scan-converting polygon meshes.
Movies	Dart (142KB).	Morph sequence (69KB), and fly-by (3.2MB), where the dart turns into the X-29 as it flies by the viewer.	X-29 (268KB).

Lion to leopard-horse morph

Description	Source	Halfway morph	Target
The lion on the left is transformed into the leopard-horse on the right; the halfway morph is shown in the middle. Both original volumes were obtained by scan-converting polygon meshes.
Movies	Lion (637KB).	Morph sequence (403KB).	Leopard-horse (925KB).

Theoretical background

Image morphing, the construction of an image sequence depicting a gradual transition between two images, has been extensively investigated. For images generated from 3D models, there is an alternative to morphing the images themselves: 3D morphing generates intermediate 3D models, the morphs, directly from the given models; the morphs are then rendered to produce an image sequence depicting the transformation. 3D morphing overcomes the following shortcomings of 2D morphing as applied to images generated from 3D models:

In 3D morphing, creating the morphs is independent of the viewing and lighting parameters. Hence, we can create a morph sequence once, and then experiment with various camera angles and lighting conditions during rendering. In 2D morphing, a new morph must be recomputed every time we wish to alter our viewpoint or the illumination of the 3D model.
2D techniques, lacking information on the model's spatial configuration, are unable to correctly handle changes in illumination and visibility. Two examples of this type of artifact are: (i) Shadows and highlights fail to match shape changes occuring in the morph. (ii) When a feature of the 3D object is not visible in the original 2D image, this feature cannot be made to appear during the morph; for example, when the singing actor needs to open her mouth during the morph, pulling her lips apart thickens the lips instead of revealing her teeth.

The models subjected to 3D morphing can be described either by geometric primitives or by volumes. Our volume morphing operates on the latter representation for reasons described in our upcoming SIGGRAPH '95 publication: Apostolos Lerios, Chase D. Garfinkle, and Marc Levoy. Feature-Based Volume Metamorphosis. This publication also contains a detailed description of our algorithm.

Acknowledgments

We would like to thank Philippe Lacroute for his help in rendering our morphs, as well as the dart to X-29 fly-by movie. The lion mesh was put together by Greg Turk as part of his research on our scanner project. We used the horse mesh courtesy of Rhythm and Hues, the color added by Greg Turk. The plastic cast of the orangutan head was lent to us by John W. Rick and was CT scanned with the help of Paul F. Hemler.