[3Dp1/3DSAp1] 3D and Hyper-realistic Displays and Applications 1

[3Dp1/3DSAp1-21] Light Field Acquisition from Focal Stack via a Deep CNN

*Yasutaka Inagaki¹, Keita Takahashi¹, Toshiaki Fujii¹ (1. Nagoya University (Japan))

キーワード：Light field, Computational photography, 3D display, Convolutional neural network

We succeeded in acquiring a dense light field from a focal stack, i.e, only a few images with different focused depth, by using a deep convolutional neural network (CNN) trained for this purpose. We validated our method through both simulative and real-camera experiments.