Light Field Convergency: Implicit Photometric Consistency on Transparent Surface

Yuta Ideguchi, Yuki Uranishi, Shunsuke Yoshimoto, Yoshihiro Kuroda, Osamu Oshiro

Abstract: This paper proposes a method for estimating the surface of transparent objects based on light field convergency. The light field convergency represents the degree of convergence of the light field at each point. The proposed method utilizes local photo consistency, which is one of characteristics of the light field convergency. Around a boundary contour, a point that is visible from viewpoints with a small convergence angle is locally consistent. In other words, the local photo consistency is implicitly maintained near the boundary contour. We use a light field camera as a camera array with small convergence angles to estimate the surface of the transparent object. Experimental results have demonstrated that the depth of the transparent object is estimated from the captured image by the light field camera.