The problem
Recovering shape from shading requires connecting lighting-dependent brightness to surface geometry. Errors in surface normals derived from an initial depth estimate can affect the recovered shape.
The research idea
A two-light photometric model optimizes initial depth, and numerical differentiation estimates surface normals. An RBF network learns the mapping from estimated to true normals on a Lambertian sphere; corrected normals are then used to optimize depth again.
What was evaluated
The publisher-deposited abstract reports comparison with a prior method on simulations and actual polyp endoscope imagery. This explainer describes the method and evaluation targets supported by that public abstract.
Considering applications
The design corrects a physics-based initial estimate with a learned model. Evaluation can examine correction under the target acquisition conditions and reconstructed-shape accuracy against an independent reference.
Original research
Recovering Polyp Shape from an Endoscope Image Using Two Light Sources
Usami, Hiroyasu, Iwahori, Yuji, Hanai, Yuki, Kijsirikul, Boonserm, Kasugai, Kunio
International Journal of Software Innovation, 2017 · Vol. 5, No. 2, pp. 33–54