Plenoptic depth estimation from multiple aliased views

With a sampled light field acquired from a plenoptic camera, several low-resolution views of the scene are available from which to infer depth. Unlike traditional multiview stereo, these views may be highly aliased due to the sparse sampling lattice in space, which can lead to reconstruction errors. We first analyse the conditions under which aliasing is a problem, and discuss the trade-offs for different parameter choices in plenoptic cameras. We then propose a method to compensate for the aliasing, whilst fusing the information from the multiple views to correctly recover depth maps. We show results on synthetic and real data, demonstrating the effectiveness of our method.