Characterization of Plenoptic Imaging Systems and Efficient Volumetric Estimation from Plenoptic Data
Plenoptic imaging is a rapidly growing field driven by the ever-declining cost of imaging systems and the promise of image focus, perspective, and depth of field manipulation during post-processing. While plenoptic systems are often limited to 2D image reconstruction and manipulation, plenoptic data and reconstruction algorithms can be extended to volumetric fields. An estimate of the imaged volume can be created by generating a stack of 2D images, but such an estimate can easily be dominated by image blur from neighboring focal planes. Tomographic algorithms have been shown to be effective in creating volumetric estimates from plenoptic data but are often prohibitively slow. The research presented here shows that the reconstruction is solvable through deconvolution. Unfortunately, the observation model is not shift-invariant. However, with appropriate transformations, the problem can be made shiftinvariant so that deconvolution is a viable solution. Utilizing the computationally efficient fast Fourier transform (FFT) allows the reconstruction to be completed quickly while producing estimates exhibiting significantly reduced blur compared to a simple focal stack. This work describes a deconvolution algorithm designed to reconstruct a 3D volume from a 2D plenoptic image. The imaging system and refocusing algorithm are characterized with respect to shift-variance in order to identify potential sources of artifacts and propose potential mitigating steps. To demonstrate the efficacy of the algorithm, experimental data is presented with comparisons of the focal stack to the reconstructed volume.