Joint Estimation of Depth, Reflectance and Illumination for Depth Refinement
Kichang Kim, Akihiko Torii, Masatoshi Okutomi · 2015
In this paper we propose a method for joint estimation of depth, reflectance and illumination from a single RGB-D image for depth refinement. This is achieved by a simple optimization based approach with smoothness constraints on depth, reflectance and illumination. We introduce an adaptively weighted local similarity constraint for reflectance, a normalized spherical-harmonic model for illumination, and an edge-aware local smoothness constraint for depth. This allows us to generate high quality depth without additional processes such as pre-training of stochastic models or image segmentation. Experimental results demonstrate that our method estimates high quality depth in comparison with ground-truth data not only for laboratory conditions but also for complex real-world scenes.