Photometric stereo is capable of high quality reconstruction
of fine shape details.
It is however prone to bias, caused by systematic error build-up, due to
imperfections in the light sources or in their calibration.
We explore possibilities of correcting the bias, using sparse
control points of known 3D location. Control points can easily be obtained
via triangulation, either by projecting a laser pattern or by adding a
camera and registering a small number of landmarks.
Previous mathematical approaches to the
incorporation of control points as constraints in
the computation of shape from normal directions lead to inadequate results.
We propose two methods for bias correction
using control points. One is based on constrained weighted-least squares
extension to shape from needle-diagram computation. The other
adds an interpolation surface to the reconstructed shape.
Experimental results demonstrate significant bias reduction,
allowing high reconstruction quality even in the presence of severe setup
calibration errors.
