Computational Imaging for Scene Understanding

6Shape from Water

Yuta ASANO¹, Yinqiang ZHANG², Ko NISHINO³, and Imari SATO⁴

¹Digital Content and Media Sciences Research Division, National Institute of Informatics, Tokyo, Japan

²Next Generation Artificial Intelligence Research Center, The University of Tokyo, Japan

³Graduate School of Informatics, Kyoto University, Japan

⁴Computational Imaging and Vision Lab, National Institute of Informatics, Tokyo, Japan

6.1. Introduction

Recovering 3D scene geometry has been one of the most important tasks in computer vision, and its methodologies have progressed significantly during the last several decades. Numerous 3D reconstruction methods have been proposed, including triangulation, time of flight and shape-from-X, where X can be shading, texture, focus or other surface or image formation properties. The fundamental but often neglected assumption of these different approaches is that the light, either actively or passively shed on the object surface including environmental illumination, can be measured unaltered between the object’s surface and the camera. Although there have been studies on shape recovery of objects in a non-air medium where this assumption does not hold (e.g. participating medium like dilute milk), their focus is on undoing adversarial optical perturbations such as scattering in order to apply the same recovery principles that were designed for objects in clear air. In other words, the medium is treated as an unwanted nuisance that violates the assumed geometry ...