Simulation of microlens array based plenoptic capture utilizing densely sampled light field

Abstract

Plenoptic cameras can capture the light field of a 3D scene in a single shot, which makes them attractive for several applications, such as depth estimation and refocusing. The difficulties in accurate calibration of available plenoptic camera designs, however, makes it also difficult to reliably assess such applications. This arises a need to have a ground-truth plenoptic data. We propose an accurate and efficient way to simulate the defocused plenoptic camera based on the geometric optics principles and the concept of densely sampled light field. In particular, we utilize the open-source computer graphics rendering software tool Blender and rely on a set of conventional 2D pinhole images of the scene captured from several viewpoints within the aperture of the main lens of the plenoptic camera. Elemental-image wise examination of plenoptic data and testing of post processing algorithms verifies the accuracy of the simulation.