All in-Focus View Synthesis from Under-Sampled Light Fields
Keita Takahashi, Akira Kubota, Takeshi Naemura · 2003
Light field rendering (LFR) is a fundamental method for generating new views from a set of pre-acquired images. We use densely-aligned cameras for the process of acquiring the set of images. In most practical cases, the density of the aligned cameras is not high enough to synthesize appropriate views. This "under-sampling" condition causes focus-like effects in the synthesized views. This paper proposes a new method for solving this problem. First, a set of differently focused views is synthesized from the undersampled set of pre-acquired images. Then, an all in-focus view is generated from the set of differently focused views. This is based on a new focus measurement algorithm specialized for light field rendering and plenoptic sampling theory. Experimental results show the effectiveness of our approach.