Depth-aware Shape from Shading

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Title: "Depth-aware Shape from Shading"

by

Mr. Linguang ZHANG

Abstract:

This thesis proposes Shape-from-Shading (SfS) from a single RGB-D image
where depth is acquired using a low-cost depth sensor. Our method works in
uncalibrated natural environment illumination, and assumes the scene is
Lambertian with constant albedo. The constant albedo assumption can be
relaxed if an additional RGB image of the same scene under different
environment lights is allowed. The key idea of our paper is to leverage
the complementary advantages of rough depth and SfS, that is, elimination
of convex/concave ambiguity and faithful reconstruction of fine geometry
details, which leads to a straightforward quadratic energy function that
optimizes a high-quality normal map and thus a detail-preserving visible
surface. We demonstrate applications of our depth-aware SfS: depth
upsampling, refinement and repair.

Date            :       5 May 2014 (Mon)

Time            :       2:00pm to 3:00pm

Venue           :       5563 (lift 27)

Advisor         :       Prof. Chi Keung TANG

2nd Reader      :       Dr. Pedro SANDER