Research collaboration

Reconstructing shape from light

Computer-vision research on recovering 3D surface shape from fewer uncalibrated images.

This research asked how to recover the three-dimensional shape of an object from images taken under changing, unknown lighting conditions—and how to do it with fewer observations.

The approach

The method combined low-rank matrix completion with integrability constraints, linking a mathematical property of the measurements to the geometry of a physically plausible surface.

What I carried forward

It was an early lesson in crossing boundaries: a useful solution can require optimization, geometry, image formation, and engineering judgment to meet in the same model.